Method performed by electronic device in communication system, electronic device, and storage medium

Abstract

The present disclosure provides a method performed by an electronic device in a communication system, an electronic device and a storage medium, relating to the fields of artificial intelligence, communication, etc. The method comprises: obtaining first training data, the first training data comprising a first sample and a first label, the first sample comprising at least one of configuration information of a base station, distribution information of the base station, distribution information of a user equipment and cell environment information, and the first label comprising statistical distribution information of at least one parameter; training a first network based on the first training data to obtain a second network; determining a second sample, the second sample comprising the same type of information as the first sample; using the second network to obtain a second label based on the second sample, and using the second sample and the second label as new training data of a third network, so that new training data can be generated based on a small amount of training data. Optionally, the above-mentioned method performed by the electronic device can be performed using an artificial intelligence model.

Description

Technical Field

[0001] This disclosure relates to the field of wireless communication technology, and more specifically, to a method, electronic device, and storage medium performed by an electronic device in a communication system. Background Technology

[0002] With the rapid development of 5G (5th Generation) and future 6G (6th Generation) communication technologies, the complexity and dynamism of wireless networks are increasing exponentially. To meet the demands for more efficient and adaptive network management, AI (Artificial Intelligence) technology, with its powerful capabilities, will become one of the means of managing wireless communication systems, enabling efficient network planning and optimization, and improving user experience. The performance of AI models is highly dependent on the quality and diversity of training data. However, wireless communication system data is characterized by high dimensionality, strong noise, high complexity, and fragmented scenarios, resulting in severely limited datasets. Therefore, constructing sufficient training data that matches wireless communication scenarios is crucial for the successful implementation of AI technology in wireless communication systems. Summary of the Invention

[0003] According to one aspect of the present disclosure, a method performed by an electronic device in a communication system is provided, the method comprising: Acquire first training data, which includes first samples and first labels. The first samples include at least one of base station configuration information, base station distribution information, user equipment distribution information, and cell environment information. The first labels include statistical distribution information of at least one parameter. Based on the first training data, the first network is trained to obtain the second network; A second sample is identified, and the information types included in the second sample are the same as those included in the first sample. Based on the second sample, the second network is used to obtain the second label, and the second sample and the second label are used as new training data for the third network.

[0004] According to another aspect of the present disclosure, an electronic device in a communication system is provided, the electronic device including a memory, a processor, and a computer program stored in the memory, the processor executing the computer program to implement the method performed by the electronic device in the communication system provided by the present disclosure.

[0005] According to another aspect of the present disclosure, a computer-readable storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements a method performed by an electronic device in a communication system provided in the present disclosure.

[0006] According to one aspect of the present disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements a method executed by an electronic device in a communication system provided in the present disclosure.

[0007] The communication system provided in this disclosure includes a method, electronic device, and storage medium executed by an electronic device. The method involves acquiring first training data, which includes a first sample and a first label. The first sample includes at least one of base station configuration information, base station distribution information, user equipment distribution information, and cell environment information. The first label includes statistical distribution information of at least one parameter. Based on the first training data, a first network is trained to obtain a second network. A second sample is determined, the information type of which is the same as that of the first sample. Based on the second sample, a second label is obtained using the second network, and the second sample and the second label are used as new training data for a third network. In this disclosure, the trained network can effectively extract key features from the wireless communication system, thus enabling the generation of a large amount of new training data based on a small amount of existing training data with similar distributions.

[0008] Furthermore, in this embodiment of the disclosure, the samples include coarse regional information such as the distribution information of base stations, the distribution information of user equipment, and / or cell environment information, rather than precise location information. This not only protects the user's privacy information, but also allows the network to learn more common features of modal information, improves the network's generalization ability, and makes the generated data more diverse and has a wider coverage. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments of this disclosure will be briefly introduced below.

[0010] Figure 1 A flowchart illustrating a method executed by an electronic device in a communication system according to an embodiment of this disclosure; Figure 2 A schematic diagram illustrating the collection of basic training data provided in an embodiment of this disclosure; Figure 3 A schematic diagram illustrating a process for training a TDG model according to an embodiment of this disclosure; Figure 4 A schematic diagram illustrating the prediction of a new parameter distribution provided in an embodiment of this disclosure; Figure 5 A schematic diagram of a framework for generating training data provided in an embodiment of this disclosure; Figure 6 This is a schematic diagram illustrating an online collection of training data provided in an embodiment of this disclosure; Figure 7 This is a schematic diagram illustrating an offline collection of training data provided in an embodiment of this disclosure; Figure 8 A schematic diagram of a fuzzing process provided in an embodiment of this disclosure; Figure 9 A schematic diagram of a CF-Net network provided in an embodiment of this disclosure; Figure 10 A schematic diagram illustrating TDG model training as provided in an embodiment of this disclosure; Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0011] The following description, with reference to the accompanying drawings, is provided to aid in a thorough understanding of the various embodiments of this disclosure as defined by the claims and their equivalents. This description includes various specific details to aid understanding but should be considered exemplary only. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the various embodiments described herein without departing from the scope and spirit of this disclosure. Furthermore, for clarity and brevity, descriptions of well-known functions and structures may be omitted.

[0012] The terms and wording used in the following description and claims are not limited to their dictionary meanings, but are merely used by the inventors to enable a clear and consistent understanding of this disclosure. Therefore, it will be apparent to those skilled in the art that the following description of various embodiments of this disclosure is for illustrative purposes only and not for limiting the purpose of this disclosure as defined in the appended claims and their equivalents.

[0013] It should be understood that the singular forms of “a,” “an,” and “the” can also include plural references unless the context clearly indicates otherwise. Thus, for example, the reference to “component surface” includes referring to one or more such surfaces. When we say that an element is “connected” or “coupled” to another element, the element can be directly connected or coupled to the other element, or it can mean that the element and the other element are connected through an intermediate element. Furthermore, the use of “connected” or “coupled” herein can include wireless connections or wireless couplings.

[0014] The terms “comprising” or “may include” refer to the presence of a corresponding disclosed function, operation, or component that may be used in the various embodiments of this disclosure, rather than limiting the presence of one or more additional functions, operations, or features. Furthermore, the terms “comprising” or “having” may be interpreted as indicating certain characteristics, numbers, steps, operations, constituent elements, components, or combinations thereof, but should not be construed as excluding the possibility of the presence of one or more other characteristics, numbers, steps, operations, constituent elements, components, or combinations thereof.

[0015] The term "or" as used in the various embodiments of this disclosure includes any of the listed terms and all combinations thereof. For example, "A or B" may include A, may include B, or may include both A and B. When describing multiple (two or more) items, if the relationship between the multiple items is not explicitly defined, the multiple items may refer to one, more, or all of the multiple items. For example, the description "parameter A includes A1, A2, A3" can be implemented as parameter A includes A1 or A2 or A3, or it can be implemented as parameter A includes at least two of the three items A1, A2, and A3.

[0016] Unless otherwise defined, all terms used in this disclosure (including technical or scientific terms) have the same meaning as understood by those skilled in the art. Common terms as defined in dictionaries are to be interpreted as having a meaning consistent with their context in the relevant technical field and should not be interpreted ideally or overly formally, unless expressly defined in this disclosure.

[0017] At least some of the functions of the device or electronic device provided in this disclosure embodiment can be implemented by an AI model, such as implementing at least one module of a plurality of modules of the device or electronic device by an AI model. AI-related functions can be executed by non-volatile memory, volatile memory, and a processor.

[0018] The processor may include one or more processors. In this case, the one or more processors may be general-purpose processors, such as central processing unit (CPU), application processor (AP), etc., or pure graphics processing unit, such as graphics processing unit (GPU), vision processing unit (VPU), and / or AI-specific processors, such as neural processing unit (NPU).

[0019] The one or more processors control the processing of input data based on predefined operating rules or artificial intelligence (AI) models stored in non-volatile and volatile memory. These predefined operating rules or AI models are provided through training or learning.

[0020] Here, "providing through learning" refers to obtaining predefined operating rules or an AI model with desired characteristics by applying a learning algorithm to multiple learning datasets. This learning can be performed within the device or electronic device itself, in which the AI ​​is executed according to the embodiment, and / or can be implemented via a separate server / system.

[0021] AI models can contain multiple neural network layers. Each layer has multiple weight values, and each layer performs neural network computations by calculating the input data of that layer (such as the computation results of the previous layer and / or the input data of the AI ​​model) and the multiple weight values ​​of the current layer. Examples of neural networks include, but are not limited to, convolutional neural networks (CNNs), deep neural networks (DNNs), recurrent neural networks (RNNs), restricted Boltzmann machines (RBMs), deep belief networks (DBNs), bidirectional recurrent deep neural networks (BRDNNs), generative adversarial networks (GANs), and deep Q-networks.

[0022] A learning algorithm is a method of training a predetermined target device (e.g., a robot) using multiple learning data sets to enable, allow, or control the target device to make determinations or predictions. Examples of such learning algorithms include, but are not limited to, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning.

[0023] According to this disclosure, at least one step in a method performed by an electronic device in a communication system, such as generating training data, can be implemented using an artificial intelligence model. The processor of the electronic device can perform preprocessing operations on the data to transform it into a form suitable for use as input to an artificial intelligence model. The artificial intelligence model can be obtained through training. Here, "obtained through training" means obtaining a predefined operating rule or artificial intelligence model configured to perform desired features (or objectives) by training a basic artificial intelligence model with multiple training data using a training algorithm.

[0024] To make the objectives, technical solutions, and advantages of this disclosure clearer, the embodiments of this disclosure will be described in further detail below with reference to the accompanying drawings.

[0025] In related technologies, to address the limitation of datasets in wireless communication systems, generative algorithms based on Generative Adversarial Networks (GANs) or Diffusion Networks are typically used to generate more training data. However, this approach has the following drawbacks: 1. Due to the inherent characteristics of GANs, gradient vanishing or gradient exploding may occur, making the training process of GANs unable to converge. 2. The sampling speed of the diffusion model is slow, usually requiring thousands of evaluation steps to extract a single sample, resulting in low data generation efficiency; 3. Training data generated based on GAN or diffusion models has a single modality, which may result in a lack of diversity and coverage in the generated data; 4. The loss function of the generator, which is widely used in GANs, makes the model vulnerable to model collapse caused by unusual optimization objectives.

[0026] In view of at least one of the above-mentioned technical problems or areas requiring improvement in the related technologies, this disclosure proposes a method, electronic device, storage medium and program product executed by an electronic device in a communication system. It can also be understood as a scheme for generating training data for AI models in wireless systems. This scheme accelerates the training and application of AI models by rapidly generating training data that matches the statistical distribution of the actual output of AI models used in wireless communication networks.

[0027] The following description of several optional embodiments illustrates the technical solutions of this disclosure and the technical effects produced by these solutions. It should be noted that the following embodiments can be referenced, learned from, or combined with each other. Identical terms, similar features, and similar implementation steps in different embodiments will not be repeated.

[0028] This disclosure provides a method executed by an electronic device in a communication system, such as... Figure 1 As shown, the method includes: Step S101: Obtain first training data. The first training data includes a first sample and a first label. The first sample includes at least one of the following: configuration information of a base station (BS or Base Transceiver Station, BTS), distribution information of base stations, distribution information of user equipment (UE), and cell environment information. The first label includes statistical distribution information of at least one parameter. In this embodiment of the disclosure, the first training data refers to the collected basic training data. The first sample in the first training data can be understood as the input to the model during model training, and the first label in the first training data can be understood as the verification of the model's output during model training.

[0029] In this embodiment of the disclosure, the first training data may be pre-collected and read and used in this step, or the first training data may be collected in real time. This embodiment of the disclosure does not limit the scope of the data.

[0030] In one example, the basic training data collected can be as follows: Figure 2As shown, this training data includes input (first sample) and labels (first label or output). The input includes base station configuration information (or system configuration information), base station distribution information, user equipment distribution information, and cell environment information. The labels include statistical distribution information for different parameters (e.g., parameter 1 to parameter n).

[0031] The type and number of parameters included in the tag can be set by those skilled in the art according to the actual situation, and the embodiments disclosed herein are not limited thereto. For example, at least one parameter may include, but is not limited to, at least one of the following: TA (Timing Advance), PMI (Precoding matrix indicator), CQI (Channel Quality Indicator), etc.

[0032] Optionally, the base station configuration information may include, but is not limited to, base station antenna line of sight, antenna height, antenna transmit power, antenna tilt angle, PRB (Physical Resource Block) utilization, uplink and downlink power control, beam configuration, neighbor cell configuration, etc.

[0033] Optionally, the distribution information of base stations and user equipment may include the distribution information of base stations and user equipment in this cell, as well as the distribution information of base stations and user equipment in neighboring cells, etc. The presentation form of the distribution information includes, but is not limited to, distribution values, distribution images, etc.

[0034] Optionally, the cell environment information may include, but is not limited to, channel information, three-dimensional map information, LOS (Line of Sight) information or NLOS (Non-Line of Sight) information of user equipment, etc.

[0035] Step S102: Based on the first training data, train the first network to obtain the second network.

[0036] In this embodiment of the disclosure, the first network refers to the network model before training, which may specifically be a preset network model. The second network refers to the network model after training. It is understood that the first network and the second network are network models with the same structure but different network parameters. In the following text, for ease of description (and to distinguish it from the third network trained using generated training data), the first network and the second network may be collectively referred to as TDG (Training Data Generation, or ATDG, Accelerator of Training Data Generation, hereinafter referred to as TDG) models.

[0037] In this step, the collected basic training data serves as the training dataset for the TDG model. The base station configuration information, base station distribution information, UE distribution information, and cell environment information from the basic training data are input into the TDG model. The statistical distribution of each parameter is used as the label to train the TDG model. Figure 3 As shown.

[0038] Optionally, the first network is trained based on the first training data. Specifically, this may include inputting the first sample into the first network for prediction, determining the difference (such as loss) between the output of the first network and the first label, adjusting the network parameters of the first network based on the difference, and repeating the process until the training termination condition is met (such as loss convergence, or training reaching a predetermined number of times or achieving a predetermined effect, etc.), to obtain the second network.

[0039] Step S103: Determine the second sample, the information type of the second sample is the same as the information type of the first sample.

[0040] In this embodiment of the disclosure, the second sample can be understood as the input of the model during the application of the TDG model. The first and second samples may have the same information content, format, parameter type, collection standard, or generation standard. That is, the second sample also includes at least one of the following: base station configuration information, base station distribution information, user equipment distribution information, and cell environment information, although the specific values ​​may differ from those of the first sample.

[0041] Optionally, the second sample can be newly collected or newly generated sample data. For example, the second sample can be collected at a different time dimension than the first sample, or the second sample can be regenerated in the same way as the first sample, such as by random generation, but is not limited to this.

[0042] Step S104: Based on the second sample, use the second network to obtain the second label, and use the second sample and the second label as new training data for the third network.

[0043] In this embodiment, a second sample is input into a second network, which then outputs a second label. It is understood that the second label output by the trained second network will have the same information content, format, parameter type, etc., as the first label. For example, the second label may include statistical distribution information of at least one parameter, which corresponds to the input second sample. Therefore, similar to the first sample data, the second sample and the second label can serve as new training data for the third network.

[0044] In other words, the trained TDG model outputs the predicted parameter distribution based on the new input information, such as... Figure 4As shown, the new input information and the predicted parameter distribution can then be used as newly generated training data.

[0045] Optionally, the second sample and the second label can be used separately as the second sample data, or alternatively, the second sample and the second label can be merged into the first sample data to form new first sample data. Those skilled in the art can choose the appropriate usage method according to the actual situation, and the embodiments disclosed herein are not limited thereto.

[0046] It is understood that, unless otherwise specified, the number of first samples, second samples, first labels, second labels, etc., appearing in this disclosure may be at least one, and the specific number shall be subject to actual implementation and shall not affect the implementation of the scheme disclosed herein.

[0047] In general, such as Figure 5 As shown, the framework for generating training data provided in this disclosure includes at least the following components: 1. Collect basic training data 2. TDG Model Training 3. Generate new training data This disclosed solution can be used for generating offline training data before deploying AI model applications in wireless communication systems, or it can be deployed on base stations for generating online training data after deploying AI model applications.

[0048] The method executed by an electronic device in a communication system provided in this disclosure allows the trained network to effectively extract key features from the wireless communication system. Therefore, it can generate a large amount of new training data based on a small amount of existing training data with similar distributions.

[0049] Furthermore, in this embodiment of the disclosure, the samples include coarse regional information such as the distribution information of base stations, the distribution information of user equipment, and / or cell environment information, rather than precise location information. This not only protects the user's privacy information, but also allows the network to learn the common features of more modal information, improves the network's generalization ability, and makes the generated data more diverse and has a wider coverage (covering more scenarios).

[0050] In this embodiment of the disclosure, the collection of basic training data may include online and / or offline methods.

[0051] Optionally, online methods may include collecting training data from base stations, such as... Figure 6 As shown, the specific operating steps may include: Step S6.1: The base station collects the parameter values ​​of the current configuration parameters of the base station online; Step S6.2: The base station collects the location and height information of base stations in its own cell, as well as the location and height information of base stations in neighboring cells (optional); Step S6.3: The base station collects user distribution snapshot information; Step S6.4: The base station collects cell environment information, including but not limited to 3D map information; Step S6.5: The base station collects the statistical distribution information of different parameters from steps S6.1 to S6.4.

[0052] Optionally, offline methods may include collecting data using a System Level Simulation (SLS) platform. Based on SLS that supports 3GPP protocols, training data can be generated by precisely configuring parameters and simulating various channel conditions, thereby enhancing the robustness of the AI ​​model. Figure 7 As shown, the specific operating steps may include: Step S7.1: Collect the configuration parameter values ​​of the base station of the SLS platform; Step S7.2: Collect the location coordinates and height information of the base station in this cell after the points are marked in the SLS platform, as well as the location coordinates and height information of the base station in the neighboring cell (optional); Step S7.3: Collect the location coordinates and height information of users in this cell after the points are marked in the SLS platform, as well as the location coordinates and height information of users in neighboring cells (optional); Step S7.4: Collect cell environment information, including but not limited to LOS information or NLOS information of user equipment; Step S7.5: Collect the statistical distribution information of different parameters output by the SLS platform in steps S7.1 to S7.4.

[0053] Furthermore, the collected training data is preprocessed to obtain the first training data.

[0054] In one optional implementation, determining the distribution information of user equipment in the first sample and / or the second sample may include: Step SA: Obtain distribution snapshot information of user equipment and / or location information of user equipment; Step SB: Preprocess the distribution snapshot information and / or location information of the user equipment to obtain the distribution information of the user equipment.

[0055] Information on user devices collected online (such as user device distribution snapshot information) and information on user devices collected offline (such as user device location information) can both be preprocessed using appropriate methods to obtain user device distribution information.

[0056] The location information of the user equipment may include the location coordinates and height information of the user equipment. Optionally, the location information of the user equipment may be preprocessed, which may include: representing the location coordinates of the user equipment as a scatter plot and dividing the height information of the user equipment into multiple height ranges to obtain a scatter plot of multiple height ranges, which is used to characterize the distribution information of the user equipment.

[0057] In other words, the distribution information of user equipment is presented in the form of scatter plots, dividing the height range into several segments and drawing the UE distribution map using several scatter plots.

[0058] As an example, such as Figure 8 As shown, the communication scenario is treated as a rectangular area, which is then divided into multiple smaller grids. This grid-based partitioning is called a grid matrix. By organizing the region into a grid matrix, the model can effectively perform structured analysis and processing of UE distribution information. BTS and UE distributions are stored in the grid matrix, and each grid contains specific information helpful for modeling and analyzing the network environment, such as the number of UEs in the cell, the average height of the UEs, the number of adjacent BTSs in the grid area, the average height of the BTSs, the average path loss, and the average shading. Environmental and channel information may also be included. By storing and processing this information in the grid matrix, the model ensures accurate context awareness, guaranteeing both privacy protection and the accuracy of the sample data.

[0059] In this embodiment of the disclosure, preprocessing the location information of the user device can also be understood as fuzzing the location information of the user device. Fuzzing allows the TDG model trained on offline data to avoid involving user location privacy information, especially location information, making it more adaptable to the field and enabling the TDG model to learn more common features, thus improving the model's generalization ability. Fuzzing involves converting precise location data into a more general or abstract form to extract features of local areas, thereby reducing the risk of exposing sensitive user information. In addition, image-based model processing can also improve the model's accuracy.

[0060] Optionally, preprocessing the distribution snapshot information of the user equipment may include: performing image processing on the distribution snapshot information using a preset image processing method.

[0061] It is understandable that, regardless of whether the training data is collected online or offline, after appropriate preprocessing, training data with the same format or type can be obtained, which is convenient for subsequent joint use.

[0062] Similarly, the base station distribution information can be preprocessed using appropriate methods to obtain the base station distribution information for both online (such as base station distribution snapshot information) and offline (such as base station location information) collected in the same way.

[0063] The location information of the base station may include the location coordinates and altitude information of the base station. Optionally, the location information of the base station may be preprocessed, which may include: representing the location coordinates of the base station as a scatter plot and dividing the altitude information of the base station into multiple altitude ranges to obtain a scatter plot of multiple altitude ranges, which is used to characterize the distribution information of the base station.

[0064] In other words, the distribution information of base stations is presented in the form of scatter plots, dividing the altitude range into several segments and drawing a BTS distribution map using several scatter plots.

[0065] Optionally, preprocessing the distribution snapshot information of the base station may include: performing image processing on the distribution snapshot information using a preset image processing method.

[0066] The beneficial effects of the preprocessing of base station information and any points not covered in detail can be found in the introduction to the UE side, and will not be repeated here.

[0067] Alternatively, the cell environment information (such as channel information) can be fuzzed. For example, a network matrix can be used for fuzzing.

[0068] Optionally, all fuzzy data (UE distribution, BTS distribution, channel, environment, etc.) and normalized system configuration can constitute the multimodal input of the TDG model.

[0069] Optionally, in addition to the preprocessing methods described above, other processing methods may include, but are not limited to, screening, filtering, transforming, enhancing, and rotating the sample data. Those skilled in the art can extend these methods according to actual circumstances, and the embodiments disclosed herein are not limited thereto.

[0070] In this embodiment of the disclosure, an optional implementation method for step S102 is provided, which may specifically include: Step S1021: Classify at least one parameter according to its dimension; Step S1022: Based on the first training data corresponding to the parameters obtained from classification, train the first network respectively to obtain the second network corresponding to the parameters of each dimension.

[0071] As an example, since TA has 32 levels, TA can be regarded as a 32-dimensional statistical value, and the number of user devices under each level can be counted as the statistical distribution information of TA.

[0072] Because CQI has 16 levels, it can be viewed as a 16-dimensional statistical value. The number of user devices under each level can be used as the statistical distribution information of CQI.

[0073] Because TA and CQI have different dimensions, they can be divided into two categories: TA and CQI. The first network is trained using the first training data corresponding to TA to obtain the second network corresponding to TA, and the first network is trained using the first training data corresponding to CQI to obtain the second network corresponding to CQI.

[0074] Alternatively, parameters with the same dimension can be grouped together.

[0075] As an example, since TA has 32 levels, it can be viewed as a 32-dimensional statistical value. The number of user devices under each level can be used as the statistical distribution information for TA. PMI is also a 32-dimensional statistical value; the number of user devices under different indices for each different RI (Rank Indication) can be used as the statistical distribution information for PMI. Therefore, TA and PMI can be considered as one class. Using the first training data corresponding to TA and PMI, a first network is trained to obtain a second network corresponding to TA and PMI.

[0076] Similarly, both uplink and downlink CQI can be considered as 16-dimensional statistical values, and the number of user devices under each level can be used as the statistical distribution information of CQI. Therefore, uplink and downlink CQI can be treated as one category. Using the first training data corresponding to the uplink and downlink CQI, the first network is trained to obtain the second network corresponding to the uplink and downlink CQI.

[0077] Another alternative implementation involves classifying at least one parameter based on the samples required for the parameter; then, based on the first training data corresponding to the classified parameters, training the first network to obtain a second network corresponding to each dimension of the parameter. The specific implementation can be similar to the implementation based on parameter dimension classification, and will not be elaborated further here.

[0078] In this embodiment of the disclosure, an optional implementation method is also provided for step S102. Specifically, it may include: training a first network based on a first training data and a first loss function to obtain a second network, wherein the first loss function is determined using KL divergence based on the statistical values ​​of each dimension of at least one parameter.

[0079] KL divergence, also known as relative entropy, is an asymmetric measure commonly used to measure the distance (similarity) between two probability distributions or two random distributions. The higher the similarity between two distributions, the smaller their KL divergence. Conversely, as the difference between two random distributions increases, their KL divergence also increases.

[0080] In the embodiments of this disclosure, since the statistical distribution information is multi-dimensional, the KL divergence can be improved to measure the similarity between the statistical values ​​of each dimension of at least one parameter in the label and the statistical values ​​of each dimension of at least one parameter predicted by the model, that is, the similarity between the two statistical distributions.

[0081] In one optional implementation, the first network can be trained for each dimension of parameters based on the corresponding first training data and the following first loss function corresponding to that dimension of parameters. The KL divergence in the first loss function can be defined as:

[0082] in, This represents the statistical distribution information of the parameters, while , , This indicates the number of cells, including the serving cell and the neighboring cell. Indicates the dimension of the parameter. Represents a statistical value for a certain dimension. , This represents the total statistical value of the community. This represents the statistical distribution information of the predicted parameters, while This represents the predicted statistical value. In this formula, It is the true statistical distribution of the parameters, and therefore is fixed. To minimize the KL divergence, we only need to minimize the latter term, i.e. and Given the cross-entropy between them, the first loss function can be expressed as:

[0083] in, This represents cell information (such as the number of cells or cell ID, including the current cell and neighboring cells), W represents the total number of user devices in the cell, and N represents the dimension of the parameter. Indicates the parameter in the first place The number of user devices in the dimension The parameter representing the prediction is in the th order. The number of user devices in the dimension.

[0084] The loss function provided in this embodiment can better reflect the differences between statistical distribution information, which helps to improve the accuracy of training the first network.

[0085] Based on at least one of the above embodiments, this disclosure provides an optional implementation method for TDG model training, and the specific operation steps may include: Step S2.1: For each parameter, classify them according to their required input and the dimensions of the parameter itself, and group parameters that require the same input and the same dimensions into one category; Step S2.2: Train the TDG model separately for parameters of different categories; Step S2.3: Input information preprocessing; When training the TDG model, the input information of the model is preprocessed; Step S2.3-1: For example, in order to enable the offline-trained TDG model to adapt to the field without involving user location privacy information, the user location and height information generated by SLS are blurred. Step S2.3-2: For example, for community environment information collected online, the community environment can be identified through image processing and other methods, and the identified information can be used as input to the TDG model; Step S2.4: Derive the loss functions for various parameters based on KL divergence; Step S2.5: Train the TDG model from step S2.2 using the corresponding input and output parameter distributions and the loss function obtained in step S2.4 to minimize the loss of the TDG model.

[0086] In this embodiment, an optional implementation of the network structure for the TDG model is also provided. Specifically, a hierarchical expansion network (CF-Net, also known as a common feature-based network) structure can be adopted, aiming to process different inputs by generating different outputs. This adaptability enables CF-Net to produce effective outputs based on different inputs. As an example, such as... Figure 9The diagram illustrates how the CF-Net network architecture is hierarchically organized to achieve its goals. The common layer in CF-Net serves as the core layer, with inputs including base station distribution information, user equipment distribution information, and base station configuration information. Based on this common layer, the CF-Net network undergoes hierarchical expansion to generate different outputs. The output generated by the CF-Net network is statistical distribution information of parameters, categorized into two distinct types based on the input data type. This hierarchical expansion allows the network to extend the input to include additional environmental information. By integrating this additional data, the network can provide more comprehensive and context-aware outputs. For example, when the input additionally includes channel information, the output corresponds to parameter type 1. Examples of such parameters include TA, etc. Similarly, when the input additionally includes environmental information, the output corresponds to parameter type 2. Examples of such parameters include PMI and CQI, etc. The CF-Net network ensures that it can effectively handle and adapt to the complexity of different inputs and can be expanded to obtain more diverse inputs to output predetermined parameter statistical distribution information.

[0087] Based on this, the embodiments of this disclosure provide an optional implementation method for TDG model training, such as... Figure 10 As shown, the training data consists of input-output pairs with known outputs. The inputs include multimodal information such as BTS / UE distribution information, system configuration, channel, and environmental information. The outputs include the statistical distribution information of the parameters. Before training the TDG model, the BTS / UE distribution information, channel information, and environmental information are fuzzed. A TDG model based on the CF-Net network is designed, and the loss function used for training is specifically designed for the statistical distribution of the parameters to improve the model's accuracy.

[0088] Based on at least one of the above embodiments, this disclosure provides an optional implementation method for generating new training data, and the specific steps may include: Step S4.1: For each type of parameter, randomly generate new inputs corresponding to that type of parameter; Step S4.2: Input the new parameters of this type into the trained TDG model; Step S4.3: The parameter distribution of the model output corresponds to the label data of the new input. The new input and output labels then constitute the newly generated training data.

[0089] Below are some application examples of generating training data with different parameters.

[0090] Application Example 1: Offline generation of TA and PMI parameter distributions (a) Offline collection of basic training data on TA and PMI from the SLS platform: Step S8.1: Collect the configuration parameter values ​​of the base station of the SLS platform, including but not limited to the base station antenna line of sight, antenna height, antenna transmit power, antenna tilt angle, PRB utilization, etc. Step S8.2: Collect the location coordinates and height information of the base station in this cell after the points are marked in the SLS platform, as well as the location coordinates and height information of the base station in the neighboring cell; Step S8.3: Collect the location coordinates and height information of users in this cell after the points are marked on the SLS platform, as well as the location coordinates and height information of users in neighboring cells; Step S8.4: Collect the user's LOS or NLOS information; Step S8.5: Collect the statistical distribution information of TA and PMI output by the SLS platform in steps S8.1 to S8.4.

[0091] (II) TDG Model Training: Step S9.1: TA is a 32-dimensional statistical value because it has 32 levels. The number of user equipment (UE) under each level is used as the statistical distribution information for TA. PMI is also a 32-dimensional statistical value. The number of UE under different indices in each RI is used as the statistical distribution information for PMI. There are a total of 4 RIs, and each RI has 8 index categories. The distributions of TA and PMI require base station configuration information, base station distribution information, UE distribution information, and UE LOS / NLOS information as inputs, and they have the same dimensions. Therefore, TA and PMI are treated as one category. Step S9.2: Design a TDG model for TA and PMI, for example, using a CNN (Convolutional Neural Networks) model. The model input includes base station configuration information, base station distribution information, user equipment distribution information, and user equipment LOS / NLOS information. The labels are 32-dimensional statistical distribution information of TA and statistical distribution information of PMI. Step S9.3: Perform fuzzy processing on the location coordinates and altitude information of the base station and user equipment: present the location coordinates and altitude information of the base station and user equipment in the form of a scatter plot. For example, divide the altitude range into several segments and use several scatter plots to draw the location map of UE and BTS; Step S9.4: The loss functions for calculating TA and PMI based on KL divergence are derived as follows:

[0092] in, This indicates the number of cells, including the cell where the base station for TA / PMI statistics is located and its neighboring cells. N represents the dimension of TA / PMI. W represents the total number of user equipment in the cell. This indicates the number of user devices in a certain dimension of the TA / PMI label. This represents the number of user devices in a given dimension of the predicted TA / PMI. Step S9.5: Train the TDG model from step S9.2 using the corresponding input and parameter distribution statistics, as well as the loss function obtained in step S9.4, to minimize the loss of the TDG model.

[0093] (III) Generation of new training data: Step S10.1: For TA and PMI, randomly generate new configuration parameter values ​​for base stations corresponding to this type of parameter, as well as the corresponding base station distribution information, user equipment distribution information, and user equipment LOS / NLOS information; Step S10.2: Input the new information into the trained TDG model; Step S10.3: The statistical distribution information of TA and PMI output by the TDG model corresponds to the label data of the new input. The new input and output label data then constitute the newly generated training data.

[0094] Application Example 2: Online Generation of CQI Distributions (a) Collect basic training data about CQI from base stations online: Step S11.1: Collect the configuration parameter values ​​of the base station online, including but not limited to the base station antenna line of sight, antenna height, antenna transmit power, antenna tilt angle, uplink and downlink power control, beam configuration, neighbor cell configuration, etc. Step S11.2: Collect the location coordinates and altitude information of the base stations in this cell, as well as the location coordinates and altitude information of the base stations in neighboring cells; Step S11.3: Collect the location coordinates and altitude distribution information of user equipment in this cell (avoiding sensitive information of specific users), as well as the corresponding downlink CQI reporting information and uplink CQI measurement information; Step S11.4: The base station collects cell environment information, including but not limited to 3D map information; Step S11.5: Collect the user distribution information and CQI statistical distribution information obtained in step S11.3.

[0095] (II) TDG Model Training: Step S12.1: Both uplink and downlink CQI are 16-dimensional statistical values ​​because there are 16 levels of CQI. The number of user devices under each level is counted as the CQI distribution. The distribution of uplink and downlink CQI requires base station configuration information, base station distribution information, user distribution information, and cell environment information as input, and they have the same dimensions; therefore, uplink and downlink CQI are treated as one category. Step S12.2: Design a TDG model for CQI, for example, using a CNN model. The model input includes base station configuration, base station distribution, user distribution, and cell environment information. The labels are 16-dimensional statistical distribution information of uplink CQI and downlink CQI. Step S12.3: Perform fuzzy processing on the location coordinates and altitude information of the base station and the location distribution information of the user equipment, for example, by presenting it in the form of a scatter plot; Step S12.4: The loss function for calculating uplink and downlink CQI based on KL divergence is derived as follows:

[0096] in, This represents the number of cells, including the cell where the base station that collects uplink and downlink CQI data is located, as well as its neighboring cells. N represents the dimension of uplink and downlink CQI. W represents the total number of user devices in the cell. This indicates the number of user devices in a certain dimension of the uplink and downlink CQI in the label. This represents the number of user devices in a certain dimension representing the predicted uplink and downlink CQI; Step S12.5: Train the TDG model from step S12.2 using the corresponding input and parameter distribution statistics, as well as the loss function obtained in step S12.4, to minimize the loss of the TDG model.

[0097] (III) Generation of new training data: Step S13.1: For uplink and downlink CQI, randomly generate new configuration parameter values ​​for base stations corresponding to this type of parameter, as well as the corresponding base station distribution information, user equipment distribution information and cell environment information; Step S13.2: Input the new information into the trained TDG model; Step S13.3: The statistical distribution information of the uplink and downlink CQI output by the TDG model corresponds to the label data of the new input. The new input and output label data then constitute the newly generated training data.

[0098] The method for generating training data proposed in this initial embodiment uses a novel artificial intelligence model instead of a GAN or diffusion model, avoiding problems such as gradient vanishing or exploding, low generation efficiency, monotonous training data modalities, and model collapse that may be caused by these models. The designed TDG model can effectively extract features from the training data, thus improving the generation speed and enabling the generation of a large amount of new training data based on a small amount of existing training data with similar distributions.

[0099] The method for generating training data proposed in this initial embodiment involves preprocessing the input data. It eliminates the need to input accurate information such as specific base station locations, user locations, and environmental information into the model. Instead, it uses only fuzzy input, including base station distribution information, user distribution information, and cell environment information, to train the TDG AI model. This allows the TDG model to learn more common features, improves the model's generalization ability, and enables the use of multimodal features, resulting in diverse and wide-ranging generated data.

[0100] The method for generating training data proposed in this initial embodiment includes a loss function design, such as a loss function for the TDG model based on KL divergence combined with multi-dimensional parameter design, which makes the statistical distribution of the generated parameters close to the real situation and improves the accuracy of generating AI models from training data.

[0101] The training data generated using the schemes of this disclosure can be used to train AI models (third networks) in a wide range of application scenarios, such as accurate channel prediction, interference identification, resource management, and network optimization. Optionally, the third network is a network that determines the base station configuration information (or scheduling information) based on at least one of the following: base station configuration information, base station distribution information, user equipment distribution information, cell environment information, and statistical distribution information of at least one parameter. In one example, if it is necessary to use the third network to adjust the base station downtilt angle during communication between the base station and the UE, the newly generated training data of this scheme can be used to train the third network. The newly generated training data can cover a wider range and is closer to the actual channel characteristics, thereby improving the training effect and performance of the third network.

[0102] This disclosure provides an electronic device in a communication system, including a transceiver configured to transmit and receive signals; and a processor coupled to the transceiver and configured to implement the steps of the foregoing method embodiments. Optionally, the node may refer to a UE, or alternatively, the node may refer to a base station, or alternatively, the node may refer to other network nodes in the wireless communication system, or alternatively, the node may also refer to a server. Detailed functional descriptions and beneficial effects can be found in the descriptions of the various method embodiments executed by the electronic device in the communication system above, and will not be repeated here. In practical applications, a UE or base station can also be understood as different network nodes.

[0103] The term "server" can refer to an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0104] Optionally, the method for generating training data provided in this disclosure can be completed collaboratively by multiple electronic devices with computing capabilities. For example, different electronic devices may each complete a portion of the steps of the methods provided in this disclosure. As an example, basic training data collection may be performed by one electronic device, TDG model training by another electronic device, and new training data generation by yet another electronic device, etc., but this is not limited to these examples.

[0105] This disclosure also provides an electronic device including a processor, and optionally, a transceiver and / or memory coupled to the processor, the processor being configured to perform the steps of the method provided in any optional embodiment of this disclosure.

[0106] Figure 11 The diagram shows a structural schematic of an electronic device to which an embodiment of the present invention applies, such as... Figure 11 As shown, Figure 11 The illustrated electronic device 4000 includes a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of this disclosure. Optionally, the electronic device may be a first network node, a second network node, or a third network node.

[0107] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with this disclosure. Processor 4001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0108] Bus 4002 may include a pathway for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 4002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 11 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0109] The memory 4003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium capable of carrying or storing computer programs and capable of being read by a computer, without limitation herein.

[0110] The memory 4003 is used to store computer programs that execute embodiments of the present disclosure, and is controlled by the processor 4001 to execute them. The processor 4001 is used to execute the computer programs stored in the memory 4003 to implement the steps shown in the foregoing method embodiments.

[0111] This disclosure provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement the steps and corresponding content of the aforementioned method embodiments.

[0112] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, can implement the steps and corresponding content of the aforementioned method embodiments.

[0113] The terms “first,” “second,” “third,” “fourth,” “1,” “2,” etc. (if present) in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in a sequence other than that shown in the figures or text.

[0114] It should be understood that although arrows indicate various operation steps in the flowcharts of the embodiments of this disclosure, the order in which these steps are implemented is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of the embodiments of this disclosure, the implementation steps in each flowchart can be executed in other orders as required. Furthermore, some or all of the steps in each flowchart may include multiple sub-steps or multiple stages based on the actual implementation scenario. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage can also be executed at different times. In scenarios where execution times differ, the execution order of these sub-steps or stages can be flexibly configured as required, and the embodiments of this disclosure do not limit this.

[0115] The above text and accompanying drawings are provided as examples only to help the reader understand this disclosure. They are not intended and should not be construed as limiting the scope of this disclosure in any way. Although certain embodiments and examples have been provided, it will be apparent to those skilled in the art, based on the content disclosed herein, that changes can be made to the illustrated embodiments and examples, and other similar implementations based on the technical concept of this disclosure can be adopted without departing from the scope of this disclosure, and these modifications and modifications are also within the protection scope of the embodiments of this disclosure.

Claims

1. A method executed by an electronic device in a communication system, characterized in that, include: Obtain first training data, which includes a first sample and a first label. The first sample includes at least one of base station configuration information, base station distribution information, user equipment distribution information, and cell environment information. The first label includes statistical distribution information of at least one parameter. Based on the first training data, the first network is trained to obtain the second network; A second sample is determined, wherein the information type included in the second sample is the same as the information type included in the first sample; Based on the second sample, the second network is used to obtain the second label, and the second sample and the second label are used as new training data for the third network.

2. The method according to claim 1, characterized in that, Determine the distribution information of user equipment in the first sample and / or the second sample, including: Obtain distribution snapshot information of user equipment and / or location information of user equipment; Preprocess the distribution snapshot information and / or location information of user equipment to obtain the distribution information of user equipment.

3. The method according to claim 2, characterized in that, The location information of the user equipment includes the location coordinates of the user equipment and the height information of the user equipment. Preprocessing of the user equipment location information includes: The location coordinates of the user equipment are represented as a scatter plot, and the height information of the user equipment is divided into multiple height ranges to obtain scatter plots of multiple height ranges, which are used to characterize the distribution information of the user equipment.

4. The method according to claim 1, characterized in that, The step of training the first network based on the first training data to obtain the second network includes: The at least one parameter is classified according to its dimension; Based on the first training data corresponding to the parameters obtained from classification, the first network is trained to obtain the second network corresponding to the parameters of each dimension.

5. The method according to claim 4, characterized in that, Based on the first training data, the first network is trained to obtain the second network, which includes: Based on the first training data and the first loss function, the first network is trained to obtain the second network, wherein the first loss function is determined using KL divergence based on the statistical values ​​of each dimension of the at least one parameter.

6. The method according to claim 1, characterized in that, The third network is a network that determines the configuration information of a base station based on at least one of the following: base station configuration information, base station distribution information, user equipment distribution information, and cell environment information.

7. The method according to claim 1, characterized in that, The at least one parameter includes at least one of the following: timing advance (TA), precoding matrix indicator (PMI), and channel quality indicator (CQI).

8. An electronic device in a communication system, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the method according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-7.

Images