Electronic device, method for wireless communication, and computer program product

Abstract

The present disclosure provides an electronic device, a method for wireless communication, a computer-readable storage medium, and a computer program product. The electronic device comprises: at least one processor; and at least one memory, comprising computer program code, wherein the at least one memory and the computer program code are configured to enable, by means of the at least one processor, the electronic device to: in an AI-assisted wireless communication scenario, acquire a channel feature vector of a current channel, the channel feature vector representing channel features in one or more dimensions; and perform lifecycle management of an AI model on the basis of the correlation between the channel feature vector of the current channel and channel feature vector distribution of the AI model, the channel feature vector distribution of the AI model being the distribution of corresponding channel feature vectors of data used by the AI model to achieve desired performance in a training or inference phase.

Description

Electronic device, method for wireless communication, and computer program product

[0001] This application claims priority to the Chinese patent application No. 202410830872.4, filed on June 25, 2024, and entitled “Electronic device, method for wireless communication, and computer program product”, the content of which is incorporated herein by reference in its entirety. TECHNICAL FIELD

[0002] The present application relates to the field of wireless communication and / or the field of artificial intelligence (AI), in particular to the lifecycle management of AI models in wireless communication based on AI model assistance. More specifically, the present application relates to an electronic device, a method for wireless communication, a computer readable storage medium, and a computer program product. BACKGROUND

[0003] The application of AI models in wireless communication is attracting more and more attention. For example, using AI to replace some traditional physical layer processes to obtain or enhance the performance of certain aspects is a hot topic in the current 3GPP RAN1. Existing use cases include AI-based beam management, AI-based positioning, AI-based channel state information (CSI) enhancement, etc., in which the following aspects of the model are mainly discussed: model training, model inference, and model monitoring, etc.

[0004] For example, in millimeter wave or higher frequency band wireless communication, beamforming technology is often needed to combat the huge path loss in communication. Beamforming technology adjusts the phase difference of a large-scale antenna array so that the energy of the transmitter (or receiver) is directed to a certain specific angle to achieve the energy gain of beamforming and improve the coverage of the wireless communication system. In general, the number of alternative beams is proportional to the number of antenna arrays, and the traditional beam scanning scheme needs to measure the reference signal received power (RSRP) of each beam to find the beam with the maximum transmit (or receive) power. Therefore, in beamforming technology, finding the optimal transmit or receive beam is the key to the problem.

[0005] With the development of AI, some researchers try to use AI methods to solve the beam prediction problem by selecting a small amount of RSRP information of the measured beams and using neural networks to predict the optimal beam. The AI-based beam prediction scheme can effectively reduce the overhead of beam measurement and greatly improve the communication efficiency. SUMMARY

[0006] A brief summary of the disclosure is presented in the following for the purpose of providing a basic understanding of certain aspects of the disclosure. It should be understood that this summary is not an exhaustive overview of the disclosure. It is not intended to determine the key or important parts of the disclosure, nor to limit the scope of the disclosure. Its purpose is only to give some concepts in a simplified form as a prelude to a more detailed description discussed later.

[0007] According to an aspect of the disclosure, an electronic device is provided, comprising: at least one processor; and at least one memory including computer program code, wherein the at least one memory and the computer program code are configured to, with the at least one processor, cause the electronic device to perform: in an AI-assisted wireless communication scenario, obtaining a channel feature vector of a current channel, the channel feature vector representing channel features in one or more dimensions; and based on a correlation between the channel feature vector of the current channel and a channel feature vector distribution of an AI model, performing lifecycle management of the AI model, the channel feature vector distribution of the AI model being a distribution of corresponding channel feature vectors of data used by the AI model to obtain a desired performance in a training or inference phase.

[0008] According to another aspect of the disclosure, a method for wireless communication is provided, comprising: in an AI-assisted wireless communication scenario, obtaining a channel feature vector of a current channel, the channel feature vector representing channel features in one or more dimensions; and based on a correlation between the channel feature vector of the current channel and a channel feature vector distribution of an AI model, performing lifecycle management of the AI model, the channel feature vector distribution of the AI model being a distribution of corresponding channel feature vectors of data used by the AI model to obtain a desired performance in a training or inference phase.

[0009] According to other aspects of the disclosure, computer program code and computer program products for implementing the above method and computer readable storage medium having the computer program code recorded thereon are also provided.

[0010] The electronic device and method according to embodiments of the present application can perform lifecycle management of an AI model based on the correlation of channel feature vectors, thereby solving the problem of matching the training scenario and the inference scenario of the model, improving the performance of the AI model and thus improving the communication quality.

[0011] These and other aspects of the present disclosure will become more apparent from the following detailed description in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0012] To further clarify the above and other advantages and features of the present disclosure, a more particular description of the application will be rendered by reference to specific embodiments thereof, which are illustrated in the appended drawings. The drawings are included to provide a description of the application and are not intended to limit the scope of the application. In the drawings:

[0013] FIG. 1 shows a schematic diagram of a conventional model switching scheme;

[0014] FIG. 2 is a functional module block diagram showing an electronic device according to an embodiment of the present application;

[0015] FIG. 3 shows a diagram of an example of a feature space and channel feature vectors when the channel feature vectors represent Doppler shift and multipath delay;

[0016] FIG. 4 shows a schematic diagram of an example where the correlation threshold in the training phase of the AI model and the correlation threshold in the inference phase are different;

[0017] FIG. 5 shows a schematic diagram of an example of a tree structure organized AI model group;

[0018] FIG. 6 shows a schematic diagram showing model structure and parameter sharing;

[0019] FIG. 7 shows a schematic diagram of an example of model expansion;

[0020] FIG. 8 shows a related flowchart of model switching;

[0021] FIG. 9 shows an example of model switching;

[0022] FIG. 10 shows a related flowchart of model update;

[0023] FIG. 11 shows a related flowchart of model switching;

[0024] FIG. 12 shows a related flowchart of model update;

[0025] FIG. 13 shows a schematic diagram of the models employed in a simulation example;

[0026] FIG. 14 shows an example of simulation results;

[0027] FIG. 15 shows a flowchart of a method for wireless communication according to an embodiment of the present application;

[0028] FIG. 16 is a block diagram illustrating a first example of a schematic configuration of an eNB or gNB to which the techniques of the present disclosure can be applied;

[0029] FIG. 17 is a block diagram illustrating a second example of a schematic configuration of an eNB or gNB to which the technology according to the present disclosure is applicable;

[0030] FIG. 18 is a block diagram illustrating an example of a schematic configuration of a smartphone to which the technology according to the present disclosure is applicable;

[0031] FIG. 19 is a block diagram illustrating an example of a schematic configuration of a car navigation device to which the technology according to the present disclosure is applicable; and

[0032] FIG. 20 is a block diagram of an example of a schematic configuration of a general-purpose personal computer in which a method and / or apparatus and / or system according to an embodiment of the present disclosure can be implemented DETAILED DESCRIPTION

[0033] In the following, exemplary embodiments of the present disclosure will be described with reference to the accompanying drawings. In the description, not all of the features of the actual implementation have been described in order to conciseness and brevity. It should be appreciated that, in the process of developing any such actual implementation, numerous implementation-specific decisions must be made in order to achieve the developer's specific goals, such as compliance with system- and business-related constraints, which will vary from one implementation to another. Moreover, it should be appreciated that, while the development work can be very laborious and time-consuming, it would be a routine undertaking for those of ordinary skill in the art having the benefit of this disclosure.

[0034] It is also noted here that, in order to avoid obscuring the present disclosure with unnecessary details, only the device structures and / or processing steps closely related to the scheme according to the present disclosure are shown in the drawings, while other details less relevant to the present disclosure are omitted.

[0035] <First Embodiment>

[0036] As described above, for example, beam prediction can be performed based on an AI model. It should be appreciated that this is only one illustrative application example, and in the following, for the ease of understanding, while the application to the beam prediction will be described as an example when necessary, the scope of application of the technology of the embodiments of the present application is not limited thereto, but can be appropriately applied to other application examples or scenarios. For example, the embodiments of the present application can be applied to other AI-assisted application scenarios in wireless communication, or to AI-assisted applications in non-wireless communication scenarios, as long as similar needs exist or similar characteristics are present.

[0037] For example, in beam prediction, AI-based methods often collect enough data sets in some scenarios and send them into a neural network for offline training, so that the output can fit the optimal beam. However, the neural network model deployed on the user equipment (or base station) side may not be the same in the inference scenario as in the training scenario, and the performance of the neural network model will be significantly degraded when it is applied to a scenario that is significantly different from the training set. Therefore, the scene needs to be distinguished to select the model suitable for the current scene (i.e., model selection or switching). As an example, FIG. 1 shows a schematic diagram of a conventional model switching scheme, in which model switching is performed according to the detected current scene. FIG. 1 shows the case of switching from model 1 for indoor low-speed scenarios to model 3 for outdoor low-speed scenarios.

[0038] In addition, in wireless communication, the real wireless environment will gradually change, for example, the movement, generation or disappearance of obstacles will bring changes in the wireless environment, and the model applied to this scenario in the past cannot meet the prediction requirements, and needs to be updated and adjusted in part to adapt to the changes in the scene (i.e., model update).

[0039] Due to the variability of wireless communication scenarios, in order to obtain better prediction performance, on the one hand, detailed scene division is needed, which requires storing a large number of AI models, and on the other hand, there are high requirements for selecting and timely switching appropriate models. Therefore, a technology for managing the life cycle of AI models is needed to ensure that appropriate AI models are applied to the current scene. The life cycle management of AI models includes, for example, the training, deployment, update, switching, expansion, deletion and other operations of AI models.

[0040] FIG. 1 shows a functional module block diagram of an electronic device 100 according to the present embodiment. As shown in FIG. 1, the electronic device 100 includes a communication unit 101 configured to obtain a channel feature vector of a current channel in an AI-assisted wireless communication scenario, the channel feature vector representing one or more dimensions of channel features; and a management unit 102 configured to perform life cycle management of AI models based on the correlation between the channel feature vector of the current channel and the channel feature vector distribution of the AI models. The channel feature vector distribution of the AI models is the distribution of the corresponding channel feature vectors of the data used by the AI models to obtain the desired performance in the training or inference stage.

[0041] The communication unit 101 and the management unit 102 can be implemented by one or more processing circuits, for example, implemented as a chip, a processor, etc., and at least one memory, for example, a RAM, a ROM, a flash memory, etc., in any form of storage device, for example, for storing computer program codes and data, etc. required by the processing circuit for processing. It should be understood that each functional unit in the electronic device shown in FIG. 1 is only a logical module according to the specific function implemented thereby, and is not intended to limit the specific implementation manner. In addition, the above description also applies to the additional functional units to be mentioned later, and will not be repeated.

[0042] The electronic device 100 can be arranged at the UE side. It should be noted that the electronic device 100 can be implemented at a chip level or also at a device level. For example, the electronic device 100 can work as the UE itself, and can further include external devices such as a memory, a transceiver (not shown in the figure), etc. The memory can be used to store programs and related data information required for the UE to implement various functions. The transceiver can include one or more communication interfaces to support communication with different devices (for example, other UEs, base stations, etc.), and the implementation form of the transceiver is not specifically limited here.

[0043] In addition, the electronic device 100 can also be arranged at the network side, such as a base station. It should be noted that the electronic device 100 can be implemented at a chip level or also at a device level. For example, the electronic device 100 can work as the base station itself, and can further include external devices such as a memory, a transceiver (not shown in the figure), etc. The memory can be used to store programs and related data information required for the base station to implement various functions. The transceiver can include one or more communication interfaces to support communication with different devices (for example, UEs, other base stations, etc.), and the implementation form of the transceiver is not specifically limited here.

[0044] In the embodiments of the present application, it is proposed to use a channel feature vector as an indication of a scenario in wireless communication. Further, the suitability of an AI model to a current scenario is evaluated by using the correlation between the channel feature vector of the current channel and the channel feature vector distribution of the AI model, thereby facilitating the life cycle management of the AI model.

[0045] The channel feature vector represents one or more dimensions of channel features. For example, the one or more dimensions of channel features can be selected from the group comprising: Doppler shift, multipath delay, signal-to-noise ratio, uplink channel fading, correlation between sub-channels in a multiple-input multiple-output (MIMO) system, speed information, location information, etc.

[0046] FIG. 3 shows a diagram of an example of a feature space and channel feature vectors, each represented as a point in the feature space, in the case that the channel feature vectors represent Doppler shift and multipath delay. The example of FIG. 3 is a diagram of two-dimensional vectors, but this is not limiting, and the channel feature vectors can also be higher-dimensional vectors.

[0047] Furthermore, when the embodiments according to the present application are applied in non-communication scenarios, the channel feature vectors will correspondingly change to feature vectors in the respective scenarios that include one or more data features that can be used to partition the model, such as in the field of visual image recognition, the data features can include RGB histograms, grayscale histograms, contour histograms, etc.

[0048] The channel feature vector distribution can include one or more of: a channel feature vector center, e.g., the average of the respective channel feature vectors; a correlation threshold of the channel feature vectors, e.g., the minimum correlation of the respective channel feature vectors with the channel feature vector center.

[0049] In other words, the channel feature vector center of the AI model is obtained by averaging the channel feature vectors of data whose scene matches the scene to which the AI model can be applied to a degree exceeding a predetermined value, and the correlation threshold is defined by the data whose channel feature vector deviates from the channel feature vector center the most (i.e., whose scene matching degree is the lowest) among the above-mentioned data. It can be understood that the correlation threshold defines the range of the scene to which the AI model can be applied. For example, when the correlation of the channel feature vector of the current channel with the channel feature vector center is higher than the correlation threshold, it indicates that the AI model is suitable for the current channel.

[0050] The correlation can be obtained, for example, by calculating the distance between the channel feature vectors, and the distance can include one of: Euclidean distance, Minkowski distance, Mahalanobis distance.

[0051] For example, as shown in FIG. 3, the left side shows the distribution of the channel feature vector of the data used when the training or inference stage of AI model 1 (referred to as model 1 for short) can obtain the expected performance, where p1 is the channel feature vector center of model 1, and r1 represents the correlation threshold of the channel feature vector of model 1 (here, shown as the Euclidean distance). The applicable range of model 1 can be the area shown in the figure with p1 as the center and r1 as the radius. The right side shows the distribution of the channel feature vector of the data used when the training or inference stage of AI model 2 (referred to as model 2 for short) can obtain the expected performance, where p2 is the channel feature vector center of model 2, and r2 represents the correlation threshold of the channel feature vector of model 2. The applicable range of model 2 can be the area shown in the figure with p2 as the center and r2 as the radius. In the example of FIG. 3, when the distance between the channel feature vector of the current channel and the channel feature vector center is less than the radius representing the correlation threshold, it means that the AI model is suitable for the current channel. The closer to the channel feature vector center, the higher the correlation.

[0052] In addition, the correlation threshold in the training stage of the AI model and the correlation threshold in the inference stage can be different, as shown in FIG. 4. That is, the applicable range of the AI model can be appropriately expanded in the inference stage.

[0053] For ease of understanding, specific examples of the life cycle management of the AI model will be given below.

[0054] For example, in the model deployment stage, the management unit 102 is configured to determine that the AI model is suitable for the current channel when the correlation between the channel feature vector of the current channel and the channel feature vector center of the AI model is greater than the correlation threshold. Still taking FIG. 3 as an example, if the channel feature vector of the current channel falls within the applicable range of model 1, it can be determined that model 1 is suitable for the current channel, and thus model 1 can be deployed.

[0055] The management unit 102 is also configured to determine to perform switching or updating of the AI model based on the correlation between the current channel feature vector of the current channel and the channel feature vector center of the currently running AI model when the performance of the currently running AI model is lower than a predetermined threshold.

[0056] For example, in a case where the correlation is less than the correlation threshold, it is determined to perform switching of the AI model; and in a case where the correlation is greater than the correlation threshold, it is determined to perform updating of the AI model. The correlation being less than the correlation threshold indicates that the current channel feature vector is seriously deviated from the channel feature vector center of the AI model, and a switching of the scene is likely to occur, and thus, it is possible to perform switching of the AI model to find an AI model suitable for the current scene. Still taking FIG. 3 as an example, if model 1 is in operation and the distance between the channel feature example of the current channel and p1 exceeds r1, it is considered that switching of the AI model is needed.

[0057] On the other hand, the correlation being greater than the correlation threshold indicates that the scene is likely not to have switched, but some factors in the scene have changed (such as movement, generation or disappearance of an obstacle, etc.), and thus, the model needs to be updated. Still taking FIG. 3 as an example, for example, the current channel feature vector is x, and still within the range where the distance from p1 does not exceed r1, it is considered that updating of model 1 is needed.

[0058] In addition, although the correlation is compared with the correlation threshold here, this is not restrictive, and for example, the correlation threshold can be appropriately increased or decreased and then compared with the correlation.

[0059] For example, the switching can include: selecting an AI model to be switched to according to the correlation between the current channel feature vector and the channel feature vector center of the AI model, and performing model performance detection until a performance requirement is met. For example, the correlation between the current channel feature vector and the channel feature vector center of the saved AI model can be compared, and the AI model with the highest correlation is selected as the AI model to be switched to; performance detection of the selected AI model is performed, if the performance requirement cannot be met, other AI models are continuously switched and performance detection is performed, otherwise the selected AI model is used.

[0060] The updating can include: updating the structure and / or parameters of the currently running AI model until the performance requirement is met. The updating can be performed through data collection and updating and model performance detection.

[0061] In addition, the management unit 102 can also be configured to, when the switching or updating is completed, determine whether to change the channel feature vector correlation threshold of the switched-to AI model or the updated AI model. For example, the management unit 102 can determine whether to change according to the performance of the AI model, for example, when the AI model applied to data with a channel feature vector less than its original correlation threshold can still obtain the expected performance, the correlation threshold can be reduced accordingly. Taking FIG. 3 as an example, the radius r1 can be increased accordingly.

[0062] On the other hand, since the scope evaluation of the scene is performed by the channel feature vector, some data of similar scenes with close channel feature vectors can be used in the training of the corresponding AI model, thereby improving the utilization rate of data samples.

[0063] In the embodiment, the channel feature vector and the distribution of the channel feature vector of the AI model are defined, and the electronic device 100 according to the embodiment determines the matching degree between the inference scene and the training scene of the AI model by comparison between the two, so as to select the appropriate AI model, thereby improving the communication quality and protecting the user privacy. In addition, the electronic device 100 can distinguish whether the decline of the model performance is caused by the model switching or updating based on the correlation between the channel feature vector and the channel feature vector center of the AI model, thereby performing more accurate model management.

[0064] <Second Embodiment>

[0065] When the number of AI models is large, a large amount of storage space will be occupied. In the embodiment, a technology for managing AI model groups in a tree structure is proposed.

[0066] For example, the AI model selection selects an AI model group organized in a tree structure, and each AI model is uniquely determined by its relative position in the tree structure, wherein AI models with close channel feature vector centers share the same parent node.

[0067] FIG. 5 shows a schematic diagram of an example of an AI model group organized in a tree structure. Each node of the tree structure except the leaf nodes represents a sub-model group, and each leaf node represents a sub-model. In the example of FIG. 5, it is shown that sub-model group 1 includes models 1 to 3, and sub-model group 2 includes models 4 to 6. As can be seen from the left side of FIG. 5, the channel feature vector centers of models 1 to 3 are close, and the channel feature vector centers of models 4 to 6 are close.

[0068] As a non-limiting example, the channel feature vector center of the sub-model group is the mean of the channel feature vector centers of each AI model in the sub-model group, and the channel feature vector correlation threshold of the sub-model group is the minimum value of the correlation of each AI model in the sub-model group with the channel feature vector center of the sub-model group. In the example of FIG. 5, the channel feature vector center of sub-model group 1 can be the mean of the channel feature vector centers of models 1 to 3, and the channel feature vector correlation threshold of sub-model group 1 can be the farthest distance from the above channel feature vector center among the channel feature vectors of models 1 to 3.

[0069] The AI models with the same parent node share the model structure and parameters of the previous part of layers of the neural network, and the shared part is stored only once in the memory, thereby effectively reducing the overhead caused by model storage. FIG. 6 shows a schematic diagram of model structure and parameter sharing, in which the model structure and parameters of the first 2 layers of model 1 and model 2 with the same parent node (sub-model group 1) are the same (i.e., the first 2 layers are the shared part), and the shared part is stored only once in the memory. Although not shown in the figure, if the sub-model group 1 further includes other models, the first 2 layers of the other models also share the shared part.

[0070] In addition, in order to further save storage space, the management unit 102 can also be configured to delete the storage content of the non-shared part of a certain AI model when the usage frequency of the AI model is lower than a predetermined frequency. When needed again, the model with the shared part of the parent node can be quickly trained on the basis of the shared network layer.

[0071] In the case of organizing the AI model group in a tree structure, the update of the AI model described in the first embodiment can include determining the layer to be updated and performing model update. For example, the management unit 102 can determine the layer to be updated according to the relevance size of the current channel feature vector and the channel feature vector center of the model. In addition, the management unit 102 can also be configured to determine whether to update the shared part of the AI model according to the performance of the AI models in the sub-model group. For example, the management unit 102 can record the performance or the number of updates of each AI model in the sub-model group within a period of time, and if it is determined to update the shared part based on the recorded performance or the number of updates, the update of the shared part is performed. It can be understood that the update of the model is similar to the training process of the model to some extent.

[0072] On the other hand, the management unit 102 can be configured to establish a new AI model in the tree structure when the relevance between the channel feature vector of the current channel and the channel feature vector center of the existing AI model is less than a predetermined threshold. The parent node of the new AI model is determined based on at least the channel feature vector of the current channel and the channel feature vector center of the existing AI model. For example, the parent node of the existing AI model corresponding to the channel feature vector center with high relevance can be selected as the parent node of the new AI model. In this way, the AI model can be expanded simply and efficiently. FIG. 7 shows a schematic diagram of one example of model expansion. When the newly detected channel feature vector of the current channel is far away from the channel feature vector center of all models and the performance of all models is poor, a new model 7 can be established in the model group, and the parent node of the model 7 is the sub-model group 2 which is close.

[0073] In addition, in the model training process, since the AI model channel feature vector centers sharing the same parent node are close, and can be applied to similar scenarios, the training samples in the similar scenarios can be fully utilized to train the shared part, thereby improving the utilization rate of the training samples.

[0074] In summary, the electronic device 100 according to the present embodiment reduces the required storage space by using a tree structure to organize the AI model. In addition, hierarchical updating of the model can also be performed.

[0075] <Third Embodiment>

[0076] When applied to auxiliary wireless communication, the AI model can be arranged at the base station side or the UE side. In the present embodiment, examples of the process of the life cycle management of the AI model when the AI model is arranged at the base station side and the UE side, respectively, will be described. It should be understood that these processes are not limiting.

[0077] First Example

[0078] In this example, the electronic device 100 is arranged at the base station side and the AI model is deployed at the base station side. The communication unit 101 is configured to receive one or more of the following information from the UE in the case where the UE detects that the performance of the currently running AI model is lower than the predetermined performance: the current channel feature vector, the correlation between the current channel feature vector and the channel feature vector center of the currently running AI model.

[0079] For example, the above information can be sent through one of the following: physical uplink control channel (PUCCH), physical uplink shared channel (PUSCH), MAC CE, radio resource control (RRC) signaling.

[0080] Specifically, the management unit 102 first performs initialization of the AI model parameters, such as setting of the channel feature vector distribution, and selection of the AI model. In addition, the management unit 102 can also perform model authentication with the UE, such as transmission of the model group including the tree structure, transmission of the capabilities of the UE such as supported channel feature vectors, transmission of the channel feature vector distribution of the model, etc. Since the AI model is deployed at the base station side, this model authentication process is optional.

[0081] During the running of the AI model, the UE can detect performance degradation, for example, the performance of the currently running AI model is lower than the predetermined performance. At this time, the UE can perform the current channel feature vector measurement or cooperate with the base station to perform the current channel feature vector measurement, and report the related information above. For example, in order to perform more accurate measurement or depending on the characteristics of the channel features to be measured, the base station can perform the measurement or cooperate with the UE to perform the measurement. In this case, the communication unit 101 can be configured to receive a request for channel feature vector measurement from the UE, and perform or cooperate with the UE to perform the channel feature vector measurement in response to the request.

[0082] For example, the communication unit 101 or the transceiver (not shown) of the electronic device 100 can send a reference signal for channel feature vector measurement; and / or indicate the use of the reference signal to the UE; and / or send an indication of starting measurement and / or reporting to the UE.

[0083] The management unit 102 can calculate the correlation between the received current channel feature vector and the channel feature vector center of the currently running AI model based on the received current channel feature vector, and determine to perform the switching or updating of the model based on the correlation; or determine to perform the switching or updating of the model based on the received correlation.

[0084] For the convenience of understanding, FIG. 8 shows the relevant flow chart of model switching, and FIG. 10 shows the relevant flow chart of model updating. The steps and flows in FIG. 8 and FIG. 10 are only illustrative and not restrictive, and can be added or omitted according to the actual situation. In FIG. 8 and FIG. 10, the base station is shown as gNB, and the user equipment is shown as UE.

[0085] As shown in FIG. 8, model authentication can be performed between the gNB and the UE. The gNB performs initialization of model parameters, such as determining the channel feature vector distribution of inference and training of the model (e.g., including the channel feature vector center and the correlation threshold, etc.). The gNB also performs selection of the model, such as according to the correlation between the current channel feature vector and the channel feature vector center of the AI model. During the running of the selected AI model, the UE detects performance degradation. At this time, the UE sends a request for channel feature vector measurement to the gNB, and the gNB performs and / or assists the UE to perform the measurement of the channel feature vector. It should be noted that the sending of the request and the assistance of the gNB are optional, and in some cases, the UE can complete the measurement of the channel feature vector by itself. Subsequently, the UE reports relevant information based on the measurement results to the gNB, which includes, for example, the measured current channel feature vector and / or the correlation between the current channel feature vector and the channel feature vector center of the currently running AI model. The gNB optionally performs the calculation of the correlation, and determines to perform model switching based on the correlation, such as the gNB finding that the correlation is less than the correlation threshold of the AI model and thus determining to perform model switching.

[0086] The gNB can determine the AI model to be switched to based on the correlation between the current channel feature vector and the channel feature vector center of other AI models, which can be calculated by the gNB or obtained from the UE. Referring to the example of FIG. 9, if the currently running model is model 4, the gNB determines that the distance between the channel feature vector center of model 4 and the current channel feature vector is greater than the applicable radius of model 4, and thus determines to perform model switching. At the same time, the gNB determines model 6 and model 5 as possible models to be switched to according to the distance between the current channel feature vector and the channel feature vector of other AI models, and performs model performance detection until the performance requirement is met, such as finally determining model 6 as the model to be switched to.

[0087] It can be seen that at this time the current channel feature vector is substantially outside the applicable radius of model 6, but its performance still meets the requirements, so the applicable radius of model 6 can be accordingly enlarged (i.e., the correlation threshold of the channel feature vector is changed). Accordingly, the step of changing the correlation threshold of the model according to the performance can also be optionally included in the flow of FIG. 8. It should be noted that the change here can be the correlation threshold of the inference of the model, or the correlation threshold of both the inference and the training of the model.

[0088] Referring to FIG. 10, it can be seen that the first half of the flow is exactly the same as the first half of the flow of FIG. 8, which is not repeated here, and only the different part is described. The gNB determines to perform the update of the model based on the correlation, for example, the gNB finds that the correlation is greater than or equal to the correlation threshold of the AI model, thereby determining to perform the update of the model, for example, the gNB can determine the level of model update and perform the model update. Then, the gNB performs data collection and model performance detection, and performs the loop until the performance requirement is met, thereby determining the final updated model. Similarly, when the scope of application of the updated model is expanded, the correlation threshold of the model can be changed. Therefore, the flow of FIG. 10 also optionally includes the step of changing the correlation threshold of the model according to the performance. Note that here the correlation threshold that can be changed is the correlation threshold of the training of the model, or the correlation threshold of both the inference and the training of the model.

[0089] Second Example

[0090] In this example, the electronic device 100 is set at the UE side, and the AI model is deployed at the UE side. In this case, the communication unit 101 is configured to perform model authentication with the base station. Similarly, the model authentication includes model group transmission of tree structure, capability of the UE such as supported channel feature vector, channel feature vector distribution transmission of the model, etc.

[0091] Similarly, the management unit 102 performs initialization of the AI model parameters such as setting of the channel feature vector distribution, and selection of the AI model.

[0092] During the running of the AI model, the UE can detect performance degradation, for example, the performance of the currently running AI model is lower than the predetermined performance. In one example, the UE can perform the measurement of the channel feature vector by itself. In another example, the communication unit 101 is configured to send a request to the base station to perform the measurement of the channel feature vector in the case where it is detected that the performance of the currently running AI model is lower than the predetermined performance. The base station can perform the measurement of the channel feature vector in cooperation with the UE in response to the request, so that more accurate measurement results can be obtained. In this case, the communication unit 101 performs the measurement of the channel feature vector in cooperation with the base station. For example, the base station can send a reference signal for the measurement of the channel feature vector; and / or indicate the use of the reference signal to the UE; and / or send an indication of starting measurement and / or reporting to the UE. The communication unit 101 receives these signals or indications accordingly, performs the measurement, etc.

[0093] In addition, the base station can also perform measurement in response to the above request, in which case the base station can also feed back the measurement result to the UE. Correspondingly, the communication unit 101 is configured to receive one or more of the following information from the base station: the current channel feature vector, the correlation between the current channel feature vector and the channel feature vector center of the currently running AI model. For example, the above information can be sent through one of the following: physical downlink control channel (PDCCH), physical downlink shared channel (PDSCH), MAC CE, radio resource control (RRC) signaling.

[0094] The management unit 102 can calculate the correlation between the current channel feature vector measured by the UE and / or received from the base station and the channel feature vector center of the currently running AI model, and determine to switch or update the model based on the correlation; or determine to switch or update the model based on the received correlation.

[0095] For ease of understanding, FIG. 11 shows a relevant flowchart of model switching, and FIG. 12 shows a relevant flowchart of model updating. The steps and flows in FIGS. 11 and 12 are only illustrative and not limiting, and can be added or omitted according to actual conditions. In FIGS. 11 and 12, the base station is shown as gNB and the user equipment is shown as UE.

[0096] As shown in FIG. 11, since the AI model is deployed on the UE side, model authentication is performed between the gNB and the UE. The UE performs initialization of the model parameters, such as determining the channel feature vector distribution (e.g. including the channel feature vector center and the correlation threshold, etc.) for inference and training of the model. The UE also performs selection of the model, for example, according to the correlation between the current channel feature vector and the channel feature vector center of the AI model. During the running of the selected AI model, the UE detects performance degradation. At this time, the UE sends a request for channel feature vector measurement to the gNB, and the gNB performs and / or assists the UE to perform measurement of the channel feature vector. It should be noted that the sending of the request and the assistance of the gNB are optional, and in some cases the UE can complete the measurement of the channel feature vector by itself.

[0097] Subsequently, the gNB feeds back to the UE relevant information based on the measurement results, which include, for example, the measured current channel feature vector and / or the correlation between the current channel feature vector and the channel feature vector center of the currently running AI model. This feedback operation is also optional and can not be performed in the case where the UE performs the measurement of the channel feature vector by itself. The UE optionally performs the calculation of the correlation and determines to perform the switching of the model based on the correlation, for example, the UE determines to perform the switching of the model when it finds that the correlation is less than the correlation threshold of the AI model. The UE can determine the AI model to be switched to based on the correlation between the current channel feature vector and the channel feature vector center of other AI models, which can be calculated by the UE or obtained from the gNB.

[0098] The step of changing the correlation threshold of the model according to the performance can also be optionally included in the flow of FIG. 11. Note that the correlation threshold changed here can be the correlation threshold of the inference of the model, or the correlation threshold of both the inference and the training of the model. The specific examples of the model switching are described in detail in the foregoing with reference to FIG. 9, which also applies here.

[0099] Referring to FIG. 12, it can be seen that the first half of the flow is exactly the same as the first half of the flow of FIG. 11, which will not be repeated here, and only the different parts will be described. The UE determines to perform the update of the model based on the correlation, for example, the UE determines to perform the update of the model when it finds that the correlation is greater than or equal to the correlation threshold of the AI model, for example, the UE can determine the level of the model update and perform the model update. Then, the UE performs data collection and model performance detection, and performs the loop until the performance requirement is met, thereby determining the final updated model. Similarly, when the applicable range of the updated model is expanded, the correlation threshold of the model can be changed. Therefore, the flow of FIG. 12 also optionally includes the step of changing the correlation threshold of the model according to the performance. Note that the correlation threshold changed here can be the correlation threshold of the training of the model, or the correlation threshold of both the inference and the training of the model.

[0100] In order to further study the beneficial technical effects of the embodiments of the present application, a simulation example is given below for reference. It should be noted that the simulation example is only for illustrative purposes and does not have a limiting effect.

[0101] In this simulation example, the transmission scenario under millimeter wave is considered, and the AI model assisted beam prediction is taken as the application scenario. The DeepMIMO O1B, O1 and I2B scenarios are used to generate channel data, wherein O1B (scenario 1) is an outdoor occluded scenario, O1 (scenario 2) is an outdoor non-occluded scenario, and I2B (scenario 3) is an indoor occluded scenario. In addition, other simulation parameters are shown in the following table:

[0102] For each scenario, 109,143 user random distribution data sample points are generated, of which 87,314 sample points are used for training and 21,829 sample points are used for testing. For each sample point, 12 uniformly distributed beam measurements are selected to measure the RSRP value, and the measurement results are sent into the neural network, and the highest probability beam is selected as the predicted beam.

[0103] The performance of four neural network models in three different scenarios is simulated, wherein model 1, model 2 and model 3 are traditional models trained for a single scenario, and model 4 is a model trained for multiple scenarios at the same time according to the embodiment, as shown in FIG. 13.

[0104] The input of all models is the RSRP value of all measured beams, and the output is the probability value of the candidate beam being the optimal beam; all models use cross-entropy as the loss function, which can be expressed as: Loss = ∑ ii y ii logp ii +(1-y ii )log(1-p ii )

[0105] where y i is an indicator variable with a value of 0 or 1, taking a value of 1 when beam i is the optimal beam, and otherwise taking a value of 0, and p i represents the probability that the i-th beam output by the neural network is the optimal beam. All models use the Adam optimizer for parameter optimization.

[0106] The difference between the four neural network model instances lies in their neural network structure and training data. Among them, models 1 to 3 all have a single-column neural network structure with a single input and a single output, model 1 has training sample data in scenario 1, model 2 has training sample data in scenario 2, and model 3 has training sample data in scenario 3; model 4 has a shared parameter neural network structure with a single input and multiple outputs (as shown in FIG. 13), and the training data is the training sample data in scenarios 1, 2 and 3. In other words, when using model 4, the technology of sharing part of the network layers of the AI model organized in a tree structure and the technology of model switching and updating based on the channel feature vector according to the embodiments of the present application can be applied.

[0107] The four models are tested in scenarios 1, 2 and 3 respectively, and the evaluation index of model performance is the prediction accuracy. Assuming that the number of samples evaluated is N1, and the number of samples in which the beam predicted by the beam is the actual optimal beam is N2, the prediction accuracy is represented as:

[0108] FIG. 14 shows simulation results, it can be seen that the performance of model 1, model 2 and model 3 is quite different in different scenarios. Model 1, 2, 3 respectively performs best in the performance of the respective training scene, and the performance is poor in the non-training scene, which shows that the model trained for a certain type of data set does not have good generalization ability to other data sets. However, model 4 shows near-optimal performance in the three scenarios, which shows that the proposed shared parameter neural network structure can well fit the data sets in the three scenarios, effectively solving the problem of matching the model inference scene and the training scene.

[0109] <Fourth embodiment>

[0110] In the above embodiments describing the processes of the electronic device for the user equipment side and the electronic device for the network side, some processes or methods are also disclosed. Hereinafter, a summary of these methods is given without repeating some details already discussed above, but it should be noted that although these methods are disclosed in the description of the processes of the electronic device for the user equipment side and the electronic device for the network side, these methods do not necessarily employ or are not necessarily performed by those components described. For example, the embodiments of the electronic device for the user equipment side and the electronic device for the network side can be partially or completely implemented using hardware and / or firmware, while the methods discussed below can be completely implemented by computer executable programs, although these methods can also employ the hardware and / or firmware of the electronic device for the user equipment side and the electronic device for the network side.

[0111] FIG. 15 shows a flowchart of a method for wireless communication according to an embodiment of the present application, the method comprising: in an AI-assisted wireless communication scenario, obtaining a channel feature vector of a current channel (S11), the channel feature vector representing one or more dimensions of channel features; and based on a correlation between the channel feature vector of the current channel and a channel feature vector distribution of an AI model, performing lifecycle management of the AI model (S12), the channel feature vector distribution of the AI model being a distribution of corresponding channel feature vectors of data used by the AI model to obtain desired performance in a training or inference phase.

[0112] For example, the channel feature vector distribution includes one or more of: a channel feature vector center, the channel feature vector center being an average of the corresponding channel feature vector; a correlation threshold of the channel feature vector, the correlation threshold of the channel feature vector being a minimum value of a correlation of the corresponding channel feature vector with the channel feature vector center.

[0113] The one or more dimensions of the channel features can be selected from a group comprising: Doppler shift, multipath delay, signal-to-noise ratio, uplink channel fading, correlation between sub-channels in a MIMO system, velocity information, location information. The correlation can be obtained by calculating the distance between the channel feature vectors, and the distance can comprise one of: Euclidean distance, Minkowski distance, Mahalanobis distance.

[0114] For example, in the model deployment phase, when the correlation between the channel feature vector of the current channel and the channel feature vector center of the AI model is greater than a correlation threshold, it is determined that the AI model is suitable for the current channel.

[0115] In the case where the performance of the currently running AI model is lower than a predetermined threshold, based on the correlation between the current channel feature vector of the current channel and the channel feature vector center of the currently running AI model, it is determined whether to perform switching or updating of the AI model.

[0116] For example, in the case where the correlation is less than a correlation threshold, it is determined to perform switching of the AI model; and in the case where the correlation is greater than a correlation threshold, it is determined to perform updating of the AI model.

[0117] The switching can comprise selecting an AI model to be switched to according to the correlation between the current channel feature vector and the channel feature vector center of the AI model, and performing model performance detection until a performance requirement is met, and the updating can comprise updating the structure and / or parameters of the currently running AI model until a performance requirement is met. The updating can be performed, for example, by data collection and updating and model performance detection.

[0118] When the switching or updating is completed, it can also be determined whether to change the channel feature vector correlation threshold of the AI model switched to or the updated AI model.

[0119] In one example, the AI models are selected from a group of AI models organized in a tree structure, each AI model being uniquely determined by its relative position in the tree structure, wherein AI models with close channel feature vector centers share the same parent node. AI models with the same parent node can share the model structure and parameters of the previous part of the neural network, and the shared part is stored only once in at least one memory.

[0120] Each node of the tree structure other than the leaf node represents a sub-model group, the channel feature vector center of the sub-model group being the mean of the channel feature vector centers of each AI model in the sub-model group, and the channel feature vector correlation threshold of the sub-model group being the minimum of the correlations between the respective channel feature vectors of each AI model in the sub-model group and the channel feature vector center of the sub-model group.

[0121] To further reduce the required storage space, the storage content of the non-shared part of the AI model can also be deleted when the usage frequency of the AI model is lower than a predetermined frequency.

[0122] In addition, when the correlation between the channel feature vector of the current channel and the channel feature vector center of the existing AI model is less than a predetermined threshold, a new AI model is established in the tree structure, and the parent node of the new AI model is determined based on at least the channel feature vector of the current channel and the channel feature vector center of the existing AI model.

[0123] For example, the updating of the AI model can include determining the layer to be updated and performing model updating. Whether to update the shared part of the AI model can be determined according to the performance of the AI model in the sub-model group.

[0124] The above method can be performed at the base station side, and the AI model is deployed at the base station side. The above method further includes receiving one or more of the following information from the UE in the case where the UE detects that the performance of the currently running AI model is lower than a predetermined performance: the current channel feature vector, the correlation between the current channel feature vector and the channel feature vector center of the currently running AI model. The information can be sent through one of the following: physical uplink control channel, physical uplink shared channel, MAC CE, and radio resource control signaling.

[0125] The above method can also include the step of model authentication with the UE.

[0126] In the case where the UE detects that the performance of the currently running AI model is lower than a predetermined performance, the above method can also include the steps of: receiving a request for channel feature vector measurement from the UE; and performing or cooperating with the UE to perform channel feature vector measurement in response to the request. For example, the above method can also include: sending a reference signal for channel feature vector measurement; and / or indicating the use of the reference signal to the UE; and / or sending an indication to the UE to start measurement and / or reporting.

[0127] The above method can also be performed at the UE side, and the AI model is deployed at the UE side. In this case, the above method further includes the step of model authentication with the base station.

[0128] In a case where it is detected that the performance of the currently running AI model is lower than the predetermined performance, the above-described method can include transmitting a request for channel feature vector measurement to the base station. For example, the UE can perform the channel feature vector measurement in cooperation with the base station; and / or receive one or more of the following information from the base station: the current channel feature vector, the correlation between the current channel feature vector and the channel feature vector center of the currently running AI model. Similarly, the above-described information is transmitted through one of the following: a physical downlink control channel, a physical downlink shared channel, a MAC CE, and radio resource control signaling.

[0129] The above-described model authentication can include information transmission of the model group of the tree structure, a supported channel feature vector, and the like.

[0130] The above-described method corresponds to the electronic device 100 of the first embodiment to the third embodiment, and the relevant specific details can be referred to the first embodiment to the third embodiment, which will not be repeated here.

[0131] The technology of the present disclosure can be applied to various products.

[0132] The electronic device 100 can be implemented as various user devices. The user device can be implemented as a mobile terminal such as a smartphone, a tablet personal computer (PC), a notebook PC, a portable game terminal, a portable / dongle type mobile router, and a digital camera device, or a vehicle-mounted terminal such as a car navigation device. The user device can also be implemented as a terminal that performs machine-to-machine (M2M) communication (also referred to as a machine type communication (MTC) terminal). In addition, the user device can be a wireless communication module (such as an integrated circuit module including a single wafer) installed on each of the above-described terminals.

[0133] For example, the electronic device 100 can be implemented as various base stations. The base station can be implemented as any type of evolved NodeB (eNB) or gNB (5G base station). The eNB includes, for example, a macro eNB and a small eNB. The small eNB can be an eNB for a small cell having a smaller coverage than a macro cell, such as a pico eNB, a micro eNB, and a home (femto) eNB. A similar situation can apply to the gNB. Alternatively, the base station can be implemented as any other type of base station, such as a NodeB and a base transceiver station (BTS). The base station can include a main body (also referred to as a base station device) configured to control wireless communication, and one or more remote radio heads (RRHs) disposed at a different place from the main body. In addition, various types of user devices can operate as a base station by temporarily or semi-persistently performing a base station function.

[0134] [Application Examples Related to Base Stations]

[0135] (First Application Example)

[0136] FIG. 16 is a block diagram illustrating a first example of a schematic configuration of an eNB or gNB to which the technology of the present disclosure can be applied. Note that the following description takes the eNB as an example, but is equally applicable to the gNB. The eNB 800 includes one or more antennas 810 and a base station device 820. The base station device 820 and each antenna 810 can be connected to each other via an RF cable.

[0137] Each of the antennas 810 includes a single or a plurality of antenna elements, such as a plurality of antenna elements included in a multiple-input multiple-output (MIMO) antenna, and is used for the base station device 820 to transmit and receive radio signals. As illustrated in FIG. 16, the eNB 800 can include a plurality of antennas 810. For example, the plurality of antennas 810 can be compatible with a plurality of frequency bands used by the eNB 800. Although FIG. 16 illustrates an example in which the eNB 800 includes a plurality of antennas 810, the eNB 800 can also include a single antenna 810.

[0138] The base station device 820 includes a controller 821, a memory 822, a network interface 823, and a wireless communication interface 825.

[0139] The controller 821 can be, for example, a CPU or a DSP, and operates various functions of a higher layer of the base station device 820. For example, the controller 821 generates data packets from data in a signal processed by the wireless communication interface 825, and transfers the generated packets via the network interface 823. The controller 821 can bundle data from a plurality of baseband processors to generate a bundled packet, and transfer the generated bundled packet. The controller 821 can have a logical function of performing control such as radio resource control, radio bearer control, mobility management, admission control, and scheduling. The control can be performed in conjunction with a nearby eNB or a core network node. The memory 822 includes a RAM and a ROM, and stores programs executed by the controller 821 and various types of control data such as a terminal list, transmission power data, and scheduling data.

[0140] The network interface 823 is a communication interface used to connect the base station device 820 to a core network 824. The controller 821 can communicate with a core network node or another eNB via the network interface 823. In this case, the eNB 800 and the core network node or the other eNB can be connected to each other by a logical interface such as an S1 interface and an X2 interface. The network interface 823 can also be a wired communication interface or a wireless communication interface for a wireless backhaul line. If the network interface 823 is a wireless communication interface, the network interface 823 can use a higher frequency band for wireless communication than a frequency band used by the wireless communication interface 825.

[0141] The wireless communication interface 825 supports any cellular communication scheme such as Long Term Evolution (LTE) and LTE-Advanced, and provides wireless connections to terminals located in a cell of the eNB 800 via the antenna 810. The wireless communication interface 825 can generally include, for example, a baseband (BB) processor 826 and an RF circuit 827. The BB processor 826 can perform, for example, encoding / decoding, modulation / demodulation, and multiplexing / demultiplexing, and perform various types of signal processing of layers (e.g., L1, medium access control (MAC), radio link control (RLC), and packet data convergence protocol (PDCP)). The BB processor 826 can have a part or all of the logical functions described above instead of the controller 821. The BB processor 826 can be a memory that stores a communication control program, or a module that includes a processor and related circuitry configured to execute the program. Updating the program can cause the functions of the BB processor 826 to change. The module can be a card or a blade that is inserted into a slot of the base station device 820. Alternatively, the module can also be a chip that is mounted on a card or a blade. Meanwhile, the RF circuit 827 can include, for example, a mixer, a filter, and an amplifier, and transmit and receive wireless signals via the antenna 810.

[0142] As illustrated in FIG. 16, the wireless communication interface 825 can include a plurality of BB processors 826. For example, the plurality of BB processors 826 can be compatible with a plurality of frequency bands used by the eNB 800. As illustrated in FIG. 16, the wireless communication interface 825 can include a plurality of RF circuits 827. For example, the plurality of RF circuits 827 can be compatible with a plurality of antenna elements. While FIG. 16 illustrates an example in which the wireless communication interface 825 includes a plurality of BB processors 826 and a plurality of RF circuits 827, the wireless communication interface 825 can also include a single BB processor 826 or a single RF circuit 827.

[0143] In the eNB 800 illustrated in FIG. 16, the communication unit 101 and the transceiver of the electronic device 100 can be implemented by the wireless communication interface 825. At least a part of the functions can also be implemented by the controller 821. For example, the controller 821 can implement the life cycle management of the AI model based on the channel feature vector and the life cycle management of the AI model group organized in a tree structure by performing the functions of the communication unit 101 and the management unit 102.

[0144] (Second Application Example)

[0145] FIG. 17 is a block diagram illustrating a second example of a schematic configuration of an eNB or gNB to which the technology of the present disclosure can be applied. Note that, similarly, the following description takes the eNB as an example, but is equally applicable to the gNB. The eNB 830 includes one or plural antennas 840, a base station device 850, and RRHs 860. The RRHs 860 and each of the antennas 840 can be connected to each other via an RF cable. The base station device 850 and the RRHs 860 can be connected to each other via a high-speed line such as an optical fiber cable.

[0146] Each of the antennas 840 includes a single or plural antenna elements (such as plural antenna elements included in a MIMO antenna) and is used for the RRH 860 to transmit and receive a radio signal. As illustrated in FIG. 17, the eNB 830 can include plural antennas 840. For example, the plural antennas 840 can be compatible with plural frequency bands used by the eNB 830. Although FIG. 17 illustrates an example in which the eNB 830 includes plural antennas 840, the eNB 830 can also include a single antenna 840.

[0147] The base station device 850 includes a controller 851, a memory 852, a network interface 853, a wireless communication interface 855, and a connection interface 857. The controller 851, the memory 852, and the network interface 853 are the same as the controller 821, the memory 822, and the network interface 823 described with reference to FIG. 16.

[0148] The wireless communication interface 855 supports any cellular communication scheme such as LTE and LTE-Advanced, and provides wireless communication to terminals located in a sector corresponding to the RRH 860 via the RRH 860 and the antennas 840. The wireless communication interface 855 can typically include, for example, a BB processor 856. The BB processor 856 is the same as the BB processor 826 described with reference to FIG. 16, except that the BB processor 856 is connected to the RF circuit 864 of the RRH 860 via the connection interface 857. As illustrated in FIG. 17, the wireless communication interface 855 can include plural BB processors 856. For example, the plural BB processors 856 can be compatible with plural frequency bands used by the eNB 830. Although FIG. 17 illustrates an example in which the wireless communication interface 855 includes plural BB processors 856, the wireless communication interface 855 can also include a single BB processor 856.

[0149] The connection interface 857 is an interface for connecting the base station device 850 (the wireless communication interface 855) to the RRH 860. The connection interface 857 can also be a communication module for communication in the above-described high-speed line for connecting the base station device 850 (the wireless communication interface 855) to the RRH 860.

[0150] The RRH 860 includes a connection interface 861 and a wireless communication interface 863.

[0151] The connection interface 861 is an interface for connecting the RRH 860 (wireless communication interface 863) to the base station apparatus 850. The connection interface 861 can also be a communication module for communication in the high-speed line described above.

[0152] The wireless communication interface 863 transmits and receives wireless signals via the antenna 840. The wireless communication interface 863 can typically include, for example, an RF circuit 864. The RF circuit 864 can include, for example, a mixer, a filter, and an amplifier, and transmits and receives wireless signals via the antenna 840. As illustrated in FIG. 17, the wireless communication interface 863 can include a plurality of RF circuits 864. For example, the plurality of RF circuits 864 can support a plurality of antenna elements. While FIG. 17 illustrates an example in which the wireless communication interface 863 includes a plurality of RF circuits 864, the wireless communication interface 863 can also include a single RF circuit 864.

[0153] In the eNB 830 illustrated in FIG. 17, the communication unit 101, the transceiver of the electronic device 100 can be implemented by the wireless communication interface 855 and / or the wireless communication interface 863. At least a part of the functions can also be implemented by the controller 851. For example, the controller 851 can implement the life cycle management of the AI model based on the channel feature vector and the life cycle management of the AI model group organized in a tree structure by executing the functions of the communication unit 101 and the management unit 102.

[0154] [Application Example Regarding User Equipment]

[0155] (First Application Example)

[0156] FIG. 18 is a block diagram illustrating an example of a schematic configuration of a smartphone 900 to which the technology of the present disclosure can be applied. The smartphone 900 includes a processor 901, a memory 902, a storage 903, an external connection interface 904, a camera 906, a sensor 907, a microphone 908, an input device 909, a display device 910, a speaker 911, a wireless communication interface 912, one or more antenna switches 915, one or more antennas 916, a bus 917, a battery 918, and an auxiliary controller 919.

[0157] The processor 901 can be, for example, a CPU or a system on chip (SoC), and controls functions of the application layer and the other layers of the smartphone 900. The memory 902 includes a RAM and a ROM, and stores data and programs executed by the processor 901. The storage 903 can include a storage medium such as a semiconductor memory and a hard disk. The external connection interface 904 is an interface for connecting an external device such as a memory card and a universal serial bus (USB) device to the smartphone 900.

[0158] The camera 906 includes an image sensor such as a charge coupled device (CCD) and a complementary metal oxide semiconductor (CMOS), and generates a captured image. The sensor 907 can include a set of sensors such as a measurement sensor, a gyro sensor, a geomagnetic sensor, and an acceleration sensor. The microphone 908 converts a sound input to the smartphone 900 into an audio signal. The input device 909 includes, for example, a touch sensor configured to detect a touch on a screen of the display device 910, a keypad, a keyboard, a button, or a switch, and receives an operation or information input from a user. The display device 910 includes a screen such as a liquid crystal display (LCD) and an organic light emitting diode (OLED) display, and displays an output image of the smartphone 900. The speaker 911 converts an audio signal output from the smartphone 900 into a sound.

[0159] The wireless communication interface 912 supports any cellular communication scheme such as LTE and LTE-Advanced, and performs wireless communication. The wireless communication interface 912 can generally include, for example, a BB processor 913 and an RF circuit 914. The BB processor 913 can perform, for example, encoding / decoding, modulation / demodulation, and multiplexing / demultiplexing, and perform various types of signal processing for wireless communication. Meanwhile, the RF circuit 914 can include, for example, a mixer, a filter, and an amplifier, and transmit and receive a wireless signal via an antenna 916. Note that, although a case in which one RF chain is connected to one antenna is shown in the figure, this is merely illustrative, and a case in which one RF chain is connected to a plurality of antennas through a plurality of phase shifters is also included. The wireless communication interface 912 can be one chip module in which the BB processor 913 and the RF circuit 914 are integrated. As shown in FIG. 18, the wireless communication interface 912 can include a plurality of BB processors 913 and a plurality of RF circuits 914. Although FIG. 18 shows an example in which the wireless communication interface 912 includes a plurality of BB processors 913 and a plurality of RF circuits 914, the wireless communication interface 912 can also include a single BB processor 913 or a single RF circuit 914.

[0160] Further, the wireless communication interface 912 can support another type of wireless communication scheme in addition to the cellular communication scheme, such as a short-range wireless communication scheme, a near field communication scheme, and a wireless local area network (LAN) scheme. In this case, the wireless communication interface 912 can include a BB processor 913 and an RF circuit 914 for each wireless communication scheme.

[0161] Each of the antenna switches 915 switches a connection destination of the antenna 916 between a plurality of circuits included in the wireless communication interface 912 (for example, circuits for different wireless communication schemes).

[0162] Each of the antennas 916 includes a single or a plurality of antenna elements (such as a plurality of antenna elements included in a MIMO antenna), and is used for the wireless communication interface 912 to transmit and receive wireless signals. As illustrated in FIG. 18, the smartphone 900 can include a plurality of antennas 916. While FIG. 18 illustrates an example in which the smartphone 900 includes a plurality of antennas 916, the smartphone 900 can also include a single antenna 916.

[0163] Further, the smartphone 900 can include an antenna 916 for each wireless communication scheme. In this case, the antenna switch 915 can be omitted from the configuration of the smartphone 900.

[0164] The bus 917 connects the processor 901, the memory 902, the storage 903, the external connection interface 904, the camera 906, the sensor 907, the microphone 908, the input device 909, the display device 910, the speaker 911, the wireless communication interface 912, and the auxiliary controller 919 to each other. The battery 918 supplies power to each block of the smartphone 900 illustrated in FIG. 18 via a feed line, which is partially illustrated as a broken line in the figure. The auxiliary controller 919 operates the minimum necessary functions of the smartphone 900, for example, in a sleep mode.

[0165] In the smartphone 900 illustrated in FIG. 18, the communication unit 101, the transceiver of the electronic device 100 can be implemented by the wireless communication interface 912. At least a part of the functions can also be implemented by the processor 901 or the auxiliary controller 919. For example, the processor 901 or the auxiliary controller 919 can implement the lifecycle management of the AI model based on the channel feature vector and the lifecycle management of the AI model group organized in a tree structure by executing the functions of the communication unit 101 and the management unit 102.

[0166] (Second Application Example)

[0167] FIG. 19 is a block diagram illustrating an example of a schematic configuration of a car navigation device 920 to which the technology of the present disclosure can be applied. The car navigation device 920 includes a processor 921, a memory 922, a global positioning system (GPS) module 924, a sensor 925, a data interface 926, a content player 927, a storage medium interface 928, an input 929, a display 930, a speaker 931, a wireless communication interface 933, one or more antenna switches 936, one or more antennas 937, and a battery 938.

[0168] The processor 921 can be, for example, a CPU or a SoC, and controls a navigation function and another function of the car navigation device 920. The memory 922 includes a RAM and a ROM, and stores data and programs executed by the processor 921.

[0169] The GPS module 924 measures a position (such as latitude, longitude, and altitude) of the car navigation device 920 using a GPS signal received from a GPS satellite. The sensor 925 can include a set of sensors such as a gyro sensor, a geomagnetic sensor, and an air pressure sensor. The data interface 926 is connected to, for example, an in-vehicle network 941 via a terminal not shown, and acquires data (such as vehicle speed data) generated by the vehicle.

[0170] The content player 927 reproduces content stored in a storage medium (such as a CD and a DVD) that is inserted into the storage medium interface 928. The input 929 includes, for example, a touch sensor configured to detect a touch on a screen of the display 930, a button, or a switch, and receives an operation or information input from a user. The display 930 includes a screen such as an LCD or an OLED display, and displays an image of the navigation function or the reproduced content. The speaker 931 outputs a sound of the navigation function or the reproduced content.

[0171] The wireless communication interface 933 supports any cellular communication scheme such as LTE and LTE-Advanced, and performs wireless communication. The wireless communication interface 933 can typically include, for example, a BB processor 934 and an RF circuit 935. The BB processor 934 can perform, for example, encoding / decoding, modulation / demodulation, and multiplexing / demultiplexing, and perform various types of signal processing for wireless communication. Meanwhile, the RF circuit 935 can include, for example, a mixer, a filter, and an amplifier, and transmit and receive wireless signals via an antenna 937. The wireless communication interface 933 can also be one chip module on which the BB processor 934 and the RF circuit 935 are integrated. As illustrated in FIG. 19, the wireless communication interface 933 can include a plurality of BB processors 934 and a plurality of RF circuits 935. While FIG. 19 illustrates an example in which the wireless communication interface 933 includes a plurality of BB processors 934 and a plurality of RF circuits 935, the wireless communication interface 933 can also include a single BB processor 934 or a single RF circuit 935.

[0172] Furthermore, the wireless communication interface 933 can support another type of wireless communication scheme in addition to the cellular communication scheme, such as a short-range wireless communication scheme, a near field communication scheme, and a wireless LAN scheme. In this case, the wireless communication interface 933 can include a BB processor 934 and an RF circuit 935 for each wireless communication scheme.

[0173] Each of the antenna switches 936 switches a connection destination of the antenna 937 between a plurality of circuits included in the wireless communication interface 933, such as circuits for different wireless communication schemes.

[0174] Each of the antennas 937 includes a single or a plurality of antenna elements, such as a plurality of antenna elements included in a MIMO antenna, and is used for the wireless communication interface 933 to transmit and receive wireless signals. As illustrated in FIG. 19, the car navigation device 920 can include a plurality of antennas 937. While FIG. 19 illustrates an example in which the car navigation device 920 includes a plurality of antennas 937, the car navigation device 920 can also include a single antenna 937.

[0175] Furthermore, the car navigation device 920 can include an antenna 937 for each wireless communication scheme. In this case, the antenna switch 936 can be omitted from the configuration of the car navigation device 920.

[0176] The battery 938 supplies power to the respective blocks of the car navigation device 920 illustrated in FIG. 19 via feed lines, which are partially illustrated as broken lines in the figure. The battery 938 accumulates power supplied from the vehicle.

[0177] In the car navigation device 920 illustrated in FIG. 19, the communication unit 101, the transceiver of the electronic device 100 can be implemented by the wireless communication interface 933. At least a part of the functions can also be implemented by the processor 921. For example, the processor 921 can implement the life cycle management of the AI model based on the channel feature vector and the life cycle management of the AI model group organized in the tree structure by executing the functions of the communication unit 101 and the management unit 102.

[0178] The technology of the present disclosure can also be implemented as an in-vehicle system (or vehicle) 940 including one or more blocks of the car navigation device 920, the in-vehicle network 941, and the vehicle module 942. The vehicle module 942 generates vehicle data such as vehicle speed, engine speed, and failure information, and outputs the generated data to the in-vehicle network 941.

[0179] The basic principles of the present disclosure are described above in connection with specific embodiments, but it should be noted that, for those skilled in the art, it can be understood that all or any steps or components of the method and apparatus of the present disclosure can be implemented in the form of hardware, firmware, software, or a combination thereof in any computing device (including processors, storage media, etc.) or network of computing devices, using the basic circuit design knowledge or basic programming skills of those skilled in the art after reading the description of the present disclosure.

[0180] Moreover, the present disclosure also proposes a program product storing machine-readable instruction codes. The instruction codes are read and executed by a machine to perform the above-mentioned method according to the embodiments of the present disclosure.

[0181] Correspondingly, the storage medium for carrying the above-mentioned program product storing machine-readable instruction codes is also included in the disclosure of the present disclosure. The storage medium includes but is not limited to floppy disks, optical disks, magneto-optical disks, memory cards, memory sticks, and the like.

[0182] In the case of implementing the present disclosure by software or firmware, the programs constituting the software are installed from the storage medium or the network to a computer (for example, the general-purpose computer 2000 shown in FIG. 20) having a dedicated hardware structure, and the computer can perform various functions and the like when various programs are installed.

[0183] In FIG. 20, a central processing unit (CPU) 2001 executes various processes according to a program stored in a read only memory (ROM) 2002 or a program loaded from a storage section 2008 to a random access memory (RAM) 2003. In the RAM 2003, data required when the CPU 2001 executes various processes and the like is also stored as necessary. The CPU 2001, the ROM 2002, and the RAM 2003 are connected to each other via a bus 2004. An input / output interface 2005 is also connected to the bus 2004.

[0184] The following components are connected to the input / output interface 2005: an input section 2006 (including a keyboard, a mouse, and the like), an output section 2007 (including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), and the like, and a speaker, and the like), a storage section 2008 (including a hard disk and the like), and a communication section 2009 (including a network interface card such as a LAN card, a modem, and the like). The communication section 2009 performs communication processing via a network such as the Internet. A drive 2010 can also be connected to the input / output interface 2005 as necessary. A removable medium 2011 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, and the like is attached to the drive 2010 as necessary, so that a computer program read therefrom is installed in the storage section 2008 as necessary.

[0185] In a case where the above series of processes are implemented by software, a program constituting the software is installed from a network such as the Internet or a storage medium such as the removable medium 2011.

[0186] It is to be understood by those skilled in the art that the storage medium is not limited to the removable medium 2011 in which the program is stored, which is distributed separately from the apparatus to provide the program to users, as illustrated in FIG. 20. Examples of the removable medium 2011 include a magnetic disk (including a floppy disk (registered trademark)), an optical disk (including a compact disc read only memory (CD-ROM) and a digital versatile disk (DVD)), a magneto-optical disk (including a mini disk (MD) (registered trademark)), and a semiconductor memory. Alternatively, the storage medium can be the ROM 2002, the hard disk contained in the storage section 2008, or the like, in which the program is stored and which is distributed together with an apparatus containing them to users.

[0187] It is also to be noted that, in the apparatus, method, and system of the present disclosure, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalents of the present disclosure. Also, the steps of executing the above series of processes can naturally be executed in time series in the order of explanation, but do not necessarily have to be executed in time series. Some steps can be executed in parallel or independently of each other.

[0188] Finally, it is to be understood that the term "including", "comprising", and variations thereof, as used in this detailed description and in the claims, are intended to be broad and encompass the terms "consisting of" and "consisting essentially of", and variations thereof. Stated otherwise, the term "including" and variations thereof are used in the detailed description and the claims in the sense of "including, but not limited to", and are intended to cover the terms "consisting of" and "consisting essentially of", and variations thereof. Furthermore, the term "consisting essentially of" and variations thereof are used in the sense of "including substantially" and are intended to cover the term "including", "including but not limited to", and variations thereof. In addition, the phrase "consisting of" and variations thereof when used herein are intended to be interpreted as "consisting of" and not "consisting essentially of" or "including" or "including but not limited to" unless otherwise indicated.

[0189] Although the embodiments of the present disclosure have been described in detail above, it should be understood that the above-described implementations are merely for the purpose of illustration and are not intended to limit the present disclosure. Various modifications and changes can be made to the above-described embodiments without departing from the spirit and scope of the present disclosure. Therefore, the scope of the present disclosure is defined only by the appended claims and their equivalents.

Claims

1. An electronic device, comprising: At least one processor; as well as At least one memory, including computer program code, wherein the at least one memory and the computer program code are configured to cause the electronic device to execute via the at least one processor: In AI-assisted wireless communication scenarios, the channel feature vector of the current channel is obtained, where the channel feature vector represents channel features of one or more dimensions; and Based on the correlation between the current channel feature vector and the channel feature vector distribution of the AI ​​model, lifecycle management of the AI ​​model is performed. The channel feature vector distribution of the AI ​​model is the distribution of the corresponding channel feature vectors of the data used by the AI ​​model to achieve the desired performance during the training or inference phase.

2. The electronic device according to claim 1, wherein, The channel feature vector distribution includes one or more of the following: a channel feature vector center, which is the average value of the corresponding channel feature vector; and a channel feature vector correlation threshold, which is the minimum correlation between the corresponding channel feature vector and the channel feature vector center.

3. The electronic device according to claim 2, wherein, The at least one memory and the computer program code are configured to cause the electronic device to execute, via the at least one processor: During the model deployment phase, if the correlation between the channel feature vector of the current channel and the center of the channel feature vector of the AI ​​model is greater than the correlation threshold, the AI ​​model is determined to be suitable for the current channel.

4. The electronic device according to claim 2, wherein, The at least one memory and the computer program code are configured to cause the electronic device to execute, via the at least one processor: If the performance of the currently running AI model is lower than a predetermined threshold, the system determines whether to switch or update the AI ​​model based on the correlation between the current channel feature vector of the current channel and the center of the channel feature vector of the currently running AI model.

5. The electronic device according to claim 4, wherein, The at least one memory and the computer program code are configured to cause the electronic device to execute, via the at least one processor: If the correlation is less than the correlation threshold, a switch to an AI model is determined; and If the correlation is greater than the correlation threshold, it is determined that the AI ​​model should be updated.

6. The electronic device according to claim 5, wherein, The switching process includes: selecting the AI ​​model to switch to based on the correlation between the current channel feature vector and the center of the AI ​​model's channel feature vector, and performing model performance testing until the performance requirements are met. The update includes updating the structure and / or parameters of the currently running AI model until performance requirements are met.

7. The electronic device according to claim 6, wherein, The update is performed through data collection and updating, as well as model performance testing.

8. The electronic device according to claim 5, wherein, The at least one memory and the computer program code are further configured to cause the electronic device to execute, via the at least one processor: When the switching or update is completed, determine whether to change the channel feature vector correlation threshold of the AI ​​model switched to or the updated AI model.

9. The electronic device according to claim 1, wherein, The channel features of one or more dimensions are selected from the group including: Doppler shift, multipath delay, signal-to-noise ratio, uplink channel fading, correlation between subchannels in a multiple-input multiple-output (MIMO) system, velocity information, and location information.

10. The electronic device according to claim 1, wherein, The correlation is obtained by calculating the distance between channel feature vectors, and the distance includes one of the following: Euclidean distance, Minkowski distance, and Mahalanobis distance.

11. The electronic device according to claim 2, wherein, The AI ​​models are selected from a group of AI models organized in a tree structure. Each AI model is uniquely determined by its relative position in the tree structure, wherein AI models with close channel feature vector centers share the same parent node.

12. The electronic device according to claim 11, wherein, AI models with the same parent node share the model structure and parameters of the first part of the neural network, and the shared part is stored only once in the at least one memory.

13. The electronic device according to claim 12, wherein, Each node in the tree structure, except for the leaf nodes, represents a sub-model group. The channel feature vector center of the sub-model group is the mean of the channel feature vector centers of each AI model in the sub-model group. The channel feature vector correlation threshold of the sub-model group is the minimum correlation between the corresponding channel feature vector of each AI model in the sub-model group and the channel feature vector center of the sub-model group.

14. The electronic device according to claim 12, wherein, The at least one memory and the computer program code are configured to cause the electronic device to execute, via the at least one processor: When the usage frequency of a certain AI model is lower than the predetermined frequency, delete the storage content of the non-shared part of the AI ​​model.

15. The electronic device according to claim 12, wherein, The at least one memory and the computer program code are configured to cause the electronic device to execute, via the at least one processor: When the correlation between the channel feature vector of the current channel and the center of the channel feature vector of the existing AI model is less than a predetermined threshold, a new AI model is established in the tree structure. The parent node of the new AI model is determined at least based on the channel feature vector of the current channel and the center of the channel feature vector of the existing AI model.

16. The electronic device according to claim 12, wherein, The update of the AI ​​model includes determining the layers to be updated and updating the model accordingly.

17. The electronic device according to claim 16, wherein, Whether to update the shared portion of the AI ​​model is determined based on the performance of the AI ​​model in the sub-model group.

18. The electronic device according to claim 4, wherein, The electronic device is located on the base station side, the AI ​​model is deployed on the base station side, and the at least one memory and the computer program code are further configured to cause the electronic device to execute via the at least one processor: If the user equipment detects that the performance of the currently running AI model is lower than the predetermined performance, one or more of the following information are received from the user equipment: the current channel feature vector, and the correlation between the current channel feature vector and the center of the channel feature vector of the currently running AI model.

19. The electronic device according to claim 18, wherein, The information is transmitted via one of the following: Physical Uplink Control Channel, Physical Uplink Shared Channel, MAC CE, or Radio Resource Control signaling.

20. The electronic device according to claim 18, wherein, The at least one memory and the computer program code are further configured to cause the electronic device to execute, via the at least one processor: Model authentication is performed between the user equipment and the user equipment.

21. The electronic device according to claim 18, wherein, The at least one memory and the computer program code are further configured to cause the electronic device to execute, via the at least one processor: Receive a request from the user equipment to perform channel feature vector measurement; and The channel feature vector measurement is performed in response to the request or in cooperation with the user equipment.

22. The electronic device according to claim 21, wherein, The at least one memory and the computer program code are further configured to cause the electronic device to execute, via the at least one processor: Transmit a reference signal for the channel feature vector measurement; and / or Instructing the user equipment of the purpose of the reference signal; and / or Send an instruction to the user equipment to begin measurement and / or reporting.

23. The electronic device according to claim 4, wherein, The electronic device is located on the user equipment side, the AI ​​model is deployed on the user equipment side, and the at least one memory and the computer program code are further configured to cause the electronic device to execute via the at least one processor: Model authentication is performed between the base station and the base station.

24. The electronic device according to claim 23, wherein, The at least one memory and the computer program code are further configured to cause the electronic device to execute, via the at least one processor: If the performance of the currently running AI model is detected to be lower than the predetermined performance, a request to perform channel feature vector measurement is sent to the base station.

25. The electronic device according to claim 23, wherein, The at least one memory and the computer program code are further configured to cause the electronic device to execute, via the at least one processor: The channel feature vector measurement is performed in cooperation with the base station; and / or Receive one or more of the following information from the base station: the current channel feature vector, and the correlation between the current channel feature vector and the center of the channel feature vector of the currently running AI model.

26. The electronic device according to claim 25, wherein, The information is transmitted via one of the following: Physical Downlink Control Channel, Physical Downlink Shared Channel, MAC CE, or Radio Resource Control signaling.

27. The electronic device according to claim 20 or claim 23, wherein, The model authentication includes information transmission of a tree-structured model group and supported channel feature vectors.

28. A method for wireless communication, comprising: In AI-assisted wireless communication scenarios, the channel feature vector of the current channel is obtained, where the channel feature vector represents channel features of one or more dimensions; and Based on the correlation between the current channel feature vector and the channel feature vector distribution of the AI ​​model, lifecycle management of the AI ​​model is performed. The channel feature vector distribution of the AI ​​model is the distribution of the corresponding channel feature vectors of the data used by the AI ​​model to achieve the desired performance during the training or inference phase.

29. A computer-readable storage medium having stored thereon computer-executable instructions that, when executed by a processor, cause the processor to perform the method according to claim 28.

30. A computer program product comprising a computer program / instructions, wherein, When the computer program / instructions are executed by the processor, they implement the steps of the method of claim 28.

Images