Space division method and device, computer equipment, storage medium and program product

Abstract

The invention relates to a space division method and device, computer equipment, a storage medium and a program product. The method comprises the following steps: acquiring beam-level signal information of each terminal; determining a communication feature corresponding to each terminal according to each piece of beam-level signal information, performing space division according to the communication feature corresponding to each terminal, and determining a space region corresponding to each terminal; the communication features corresponding to all the terminals in each space area are consistent. By using the space division method provided by the invention, the communication characteristics in the divided space regions can be ensured to be consistent.

Description

Technical Field

[0001] The present application relates to the field of wireless communication technology, and in particular to a space division method, apparatus, computer equipment, storage medium and program product. Background Art

[0002] In wireless communication networks, in order to reduce the complexity of calculation and storage, it is necessary to transform the continuous optimization form problem of the entire network space into a discrete optimization form problem through spatial discretization, that is, to perform spatial division or rasterization.

[0003] In traditional technology, the geographic space is usually divided into multiple spatial regions, namely grids, and the terminals are assigned to corresponding grids according to their geographic location information to complete the division of space.

[0004] However, similar geographical locations do not necessarily represent similar communication characteristics. The communication characteristics within a spatial area divided based on geographical location information may be inconsistent. Summary of the invention

[0005] Based on this, it is necessary to provide a space division method, device, computer equipment, storage medium and program product that can ensure consistent communication characteristics within the divided space areas in response to the above technical problems.

[0006] In a first aspect, the present application provides a space division method, the method comprising:

[0007] Obtaining beam-level signal information of each terminal;

[0008] The communication characteristics corresponding to each terminal are determined according to the signal information of each beam level, and space division is performed according to the communication characteristics corresponding to each terminal to determine the space area corresponding to each terminal; the communication characteristics corresponding to all terminals in each space area are consistent.

[0009] In one embodiment, determining the communication characteristics corresponding to each terminal according to each beam-level signal information, and performing spatial division according to the communication characteristics corresponding to each terminal to determine the spatial area corresponding to each terminal includes:

[0010] Analyze the signal information of each beam level to determine the communication characteristics corresponding to each terminal and the communication characteristics of the regional center;

[0011] The space is divided according to the communication characteristics corresponding to each terminal and the communication characteristics of the regional center, and the space area corresponding to each terminal is determined.

[0012] In one embodiment, analyzing each beam-level signal information to determine the communication characteristics corresponding to each terminal and the communication characteristics of the regional center includes:

[0013] Input each beam-level signal information into the target encoding subnetwork in the target deep learning model to obtain the communication characteristics corresponding to each terminal;

[0014] The communication features corresponding to each terminal are input into the target query subnetwork in the target deep learning model to obtain the regional center communication features.

[0015] In one embodiment, analyzing each beam-level signal information to determine the communication characteristics corresponding to each terminal and the communication characteristics of the regional center includes:

[0016] The spatial partitioning objective function is constructed according to the signal information of each beam level, the communication characteristics corresponding to each terminal, and the communication characteristics of the regional center;

[0017] Solve the space partition objective function to obtain the mapping sub-function and regional center communication characteristics;

[0018] According to the mapping sub-function and the signal information of each beam level, the communication characteristics corresponding to each terminal are determined.

[0019] In one embodiment, the training method of the target deep learning model includes:

[0020] Obtaining beam-level signal information samples;

[0021] Inputting beam-level signal information samples into the initial deep learning model to obtain reconstructed beam-level signal information;

[0022] Determining a reconstruction loss based on the reconstructed beam-level signal information and the beam-level signal information sample;

[0023] The initial deep learning model is trained according to the reconstruction loss to obtain the target deep learning model.

[0024] In one embodiment, the beam-level signal information sample is input into an initial deep learning model to obtain reconstructed beam-level signal information, including:

[0025] In the case where there is missing data in the beam-level signal information samples, determining a missing vector;

[0026] Completing the missing data in the beam-level signal information sample to obtain a completed beam-level signal information sample;

[0027] Input the completed beam-level signal information samples into the initial deep learning model to obtain reconstructed beam-level signal information;

[0028] Determining a reconstruction loss according to the reconstructed beam-level signal information and the beam-level signal information sample includes:

[0029] The reconstruction loss is determined according to the reconstructed beam-level signal information, the completed beam-level signal information samples and the missing vector.

[0030] In one embodiment, the initial deep learning model includes a first initial encoding subnetwork, a first initial query subnetwork, and a first initial decoding subnetwork, and the beam-level signal information sample is input into the initial deep learning model to obtain reconstructed beam-level signal information, including:

[0031] Inputting the beam-level signal information sample into the first initial coding subnetwork to obtain a communication feature sample;

[0032] Inputting the communication feature sample into the first initial query subnetwork to obtain the regional center communication feature sample;

[0033] The regional center communication feature samples are input into the first initial decoding sub-network to obtain the reconstructed beam-level signal information; the first initial decoding sub-network includes a neural network or a physical model.

[0034] In one embodiment, the initial deep learning model is trained according to the reconstruction loss to obtain the target deep learning model, including:

[0035] Determine the discretization loss according to the communication feature samples and the regional center communication feature samples;

[0036] The initial deep learning model is trained according to the reconstruction loss and the discretization loss, and the first initial encoding subnetwork and the first initial query subnetwork in the trained initial deep learning model are determined as the target deep learning model.

[0037] In one embodiment, training the initial deep learning model according to the reconstruction loss and the discretization loss includes:

[0038] Based on the straight-through estimation method or the soft vector quantization method, the initial deep learning model is trained according to the reconstruction loss and the discretization loss.

[0039] In one embodiment, the initial deep learning model includes a second initial encoding subnetwork, a second initial query subnetwork, and a second initial decoding subnetwork, and the beam-level signal information sample is input into the initial deep learning model to obtain reconstructed beam-level signal information, including:

[0040] Inputting the beam-level signal information sample into the second initial encoding subnetwork to obtain a dimension-reduced vector;

[0041] Input the dimension-reduced vector into the second initial query sub-network to obtain a regional center communication feature sample;

[0042] The regional center communication feature samples are input into the second initial decoding sub-network to obtain the reconstructed beam-level signal information.

[0043] In one of the embodiments, the beam-level signal information sample includes geographic location information of the terminal, and the dimension of the reduced-dimensionality vector is the same as the dimension of the geographic location information.

[0044] In a second aspect, an embodiment of the present application provides a space division device, the device comprising:

[0045] An acquisition module, used to acquire beam-level signal information of each terminal;

[0046] The division module is used to determine the communication characteristics corresponding to each terminal according to the signal information of each beam level, and to perform spatial division according to the communication characteristics corresponding to each terminal to determine the spatial area corresponding to each terminal; the communication characteristics corresponding to all terminals in each spatial area are consistent.

[0047] In a third aspect, an embodiment of the present application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method provided in the first aspect when executing the computer program.

[0048] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method provided in the first aspect are implemented.

[0049] In a fifth aspect, an embodiment of the present application further provides a computer program product, including a computer program, which implements the steps of the method provided in the first aspect when the computer program is executed by a processor.

[0050] The above-mentioned space division method, device, computer equipment, storage medium and program product, the method obtains the beam-level signal information of each terminal; determines the communication characteristics corresponding to each terminal according to each beam-level signal information, and performs space division according to the communication characteristics corresponding to each terminal to determine the space area corresponding to each terminal; the communication characteristics corresponding to all terminals in each space area are consistent. In this embodiment, when performing space division, the division is performed according to the communication characteristics corresponding to each terminal determined by each beam-level signal information, so that the communication characteristics of each terminal in each divided space area can be guaranteed to be consistent, thereby improving the benefits of network optimization and wireless resource management in each space area. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] FIG1 is an application environment diagram of a space partitioning method in one embodiment;

[0052] FIG2 is a schematic diagram of a flow chart of steps of a space division method in one embodiment;

[0053] FIG3 is a schematic flow chart of steps of a space division method according to another embodiment;

[0054] FIG4 is a schematic flow chart of steps of a space division method according to another embodiment;

[0055] FIG5 is a schematic flow chart of steps of a space division method according to another embodiment;

[0056] FIG6 is a schematic flow chart of steps of a space division method according to another embodiment;

[0057] FIG7 is a schematic flow chart of steps of a space division method according to another embodiment;

[0058] FIG8 is a schematic diagram of the structure of an initial deep learning model in one embodiment;

[0059] FIG9 is a schematic flow chart of steps of a space division method according to another embodiment;

[0060] FIG10 is a schematic flow chart of steps of a space division method according to another embodiment;

[0061] FIG11 is a schematic diagram of the structure of an initial deep learning model in another embodiment;

[0062] FIG12 is a schematic flow chart of steps of a space division method according to another embodiment;

[0063] FIG13 is a schematic flow chart of steps of a space division method according to another embodiment;

[0064] FIG14 is a schematic diagram of the structure of a space dividing device in one embodiment;

[0065] FIG. 15 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0067] The serial numbers assigned to the components in this article, such as "first", "second", etc., are only used to distinguish the objects described and do not have any order or technical meaning.

[0068] First of all, before specifically introducing the technical solutions of the embodiments disclosed in the present application, the background technology or technical evolution context on which the embodiments of the present application are based is introduced. In the design and optimization of modern wireless communication networks, effectively managing and predicting wireless communication network resources and their distribution is the key to improving network performance. It is crucial to understand and predict the distribution and development trend of wireless communication network resources in a specific area, which not only affects load balancing and service quality, but also is the basis for network parameter optimization. In large-scale network optimization, in order to reduce the complexity of calculation and storage, it is necessary to convert the continuous optimization form problem of the entire space into a discrete optimization form problem through spatial discretization, that is, to perform spatial division or rasterization. An ideal spatial region usually represents an area with similar communication characteristics. In network optimization problems, it can be approximately considered that the communication conditions within the spatial region are different, and they can be regarded as a whole during the optimization process, which can reduce the complexity of optimization.

[0069] In traditional technology, there are two methods for dividing network space. The first method is to obtain the geographical location information of the terminal through a road driving test (Drive Test, DT), and perform effective space division based on the geographical location information. However, this method requires the simultaneous collection of the geographical location information of the wireless terminal under the coverage area of ​​the base station. Due to the limitations of privacy protection policies or terminal settings, for example, user devices (mobile phones, computers, car computers, etc.), usually do not directly report their geographical location information. Although communication samples including geographical location information can be obtained through DT, the collection cost is high and the coverage is limited to public road areas. In addition, similar geographical locations do not necessarily represent similar communication characteristics, which may lead to inaccurate solution of communication characteristics of the divided spatial area, and reduced benefits of network optimization and wireless resource management. The second method is to use the beam-level reference signal received power (RSRP) in the measurement report (MR) data as coordinates to establish a high-dimensional beam space, and then cluster the RSRP data. Each clustered class is equivalent to a discrete beam space area, and the samples in the same spatial area have similar RSRP characteristics. However, similar RSRP does not mean similar communication characteristics. The communication conditions of the terminals corresponding to similar RSRP data may be completely different. If terminals with different communication characteristics but the same RSRP are divided into a spatial area and regarded as similar in the network optimization problem, it will bring inevitable deviation effects to the network optimization problem. In this regard, the present application provides a spatial partitioning method that can ensure that the communication characteristics in the same spatial area after division are similar.

[0070] The space division method provided in the present application can be applied to an application environment as shown in Figure 1, which includes multiple terminals 102 and base stations 104, and the base stations 104 are connected to each other in communication. The terminal 102 can be, but is not limited to, a mobile phone, a laptop computer, an industrial computer, a tablet computer, and a wearable device. The base station 104 includes a base transceiver station and a base station controller.

[0071] The technical solution of the present application and how the technical solution of the present application solves the technical problem are described in detail below with specific embodiments.

[0072] In one embodiment, as shown in FIG2 , a space division method is provided. This embodiment is illustrated by applying the method to a base station controller in a base station. In this embodiment, the method includes the following steps:

[0073] Step 200: Obtain beam-level signal information of each terminal.

[0074] The base transceiver station in the base station can receive information of each terminal in the coverage area of ​​the base station, and the base transceiver station can send the received information to the base station controller. After receiving the information, the base station controller obtains the beam-level signal information of each terminal from the information. The beam-level signal information of each terminal can be the RSRP sent by the terminal to the base station.

[0075] In an optional embodiment, the information of each terminal received by the base transceiver station may include drive test (DT) data or measurement report (MR) data. The drive test data is shown in Table 1, and the measurement report data is shown in Table 2. Among them, UE1 (User Equipment), UE2... represent multiple terminals, CSI-RS (Channel State Information-Reference Signal) RSRP represents the receiving power of the channel state information reference signal. SSB (Synchronization Signal / PBCH) RSRP represents the receiving power of the synchronization signal. SRS (Sounding Reference Signal) RSRP represents the receiving power of the uplink sounding reference signal.

[0076] Table 1

[0077]

[0078] Table 2

[0079]

[0080] Step 210: determine the communication characteristics corresponding to each terminal according to the signal information of each beam level, and perform spatial division according to the communication characteristics corresponding to each terminal to determine the spatial area corresponding to each terminal; the communication characteristics corresponding to all terminals in each spatial area are consistent.

[0081] The communication characteristics corresponding to each terminal may be the geographical location information, angular power spectrum, spectrum efficiency, number of channels, etc. of each terminal. This embodiment does not limit the specific type of communication characteristics, and the user can set them according to actual needs.

[0082] After acquiring the beam-level signal information of each terminal, the base station controller analyzes the beam-level signal information of each terminal to determine the communication characteristics corresponding to each terminal. If the communication characteristic is an angular power spectrum, the base station controller can extract RSRP data from the acquired beam-level signal information and calculate the angular power spectrum of each terminal based on the RSRP data. If the communication characteristic is spectral efficiency, the base station controller can determine the spectral efficiency of each terminal from the acquired beam-level signal information. This embodiment does not limit the specific method of determining the communication characteristics corresponding to each terminal according to each beam-level signal information, as long as its function can be achieved.

[0083] After determining the communication characteristics corresponding to each terminal, the base station controller performs spatial division based on the communication characteristics of each terminal, and determines the spatial area corresponding to each terminal, so that the communication characteristics of all terminals in each spatial area are consistent. In other words, when performing spatial division based on the communication characteristics of each terminal, terminals with consistent communication characteristics are divided into the same spatial area. This embodiment does not limit the specific method of performing spatial division according to the communication characteristics corresponding to each terminal, as long as it can ensure that the communication characteristics of all terminals in each divided spatial area are consistent.

[0084] In an optional embodiment, the base station controller may perform space division by calculating the similarity between the communication features corresponding to the terminals.

[0085] The spatial division method provided in the embodiment of the present application obtains the beam-level signal information of each terminal; determines the communication characteristics corresponding to each terminal according to each beam-level signal information, and performs spatial division according to the communication characteristics corresponding to each terminal to determine the spatial area corresponding to each terminal; the communication characteristics corresponding to all terminals in each spatial area are consistent. In this embodiment, when performing spatial division, the division is performed according to the communication characteristics corresponding to each terminal determined by each beam-level signal information, so that the communication characteristics of each terminal in each divided spatial area can be guaranteed to be consistent, thereby improving the benefits of network optimization and wireless resource management in each spatial area.

[0086] In one embodiment, as shown in FIG3 , a method for determining communication characteristics corresponding to each terminal according to each beam-level signal information, performing spatial division according to the communication characteristics corresponding to each terminal, and determining a spatial region corresponding to each terminal is implemented, and the steps of the implementation method include:

[0087] Step 300: Analyze the signal information of each beam level to determine the communication characteristics corresponding to each terminal and the communication characteristics of the regional center.

[0088] After obtaining the beam-level signal information of each terminal, the base station controller analyzes the beam-level signal information to determine the communication characteristics and regional center communication characteristics corresponding to each terminal. For the description of determining the communication characteristics corresponding to each terminal, reference can be made to the specific description in the above embodiment, which will not be repeated here. The regional center communication characteristics refer to the communication characteristics of the regional center of the divided spatial area. The number of regional center communication characteristics can include multiple ones, which can be set according to the needs of the user. The regional center communication characteristics can be any one or more of the communication characteristics corresponding to each terminal, can be determined based on the communication characteristics corresponding to each terminal, or can be pre-set by the user.

[0089] Step 310: Perform spatial division according to the communication characteristics corresponding to each terminal and the communication characteristics of the regional center to determine the spatial area corresponding to each terminal.

[0090] After determining the communication characteristics corresponding to each terminal and the communication characteristics of the regional center, the base station controller uses a clustering method to perform spatial division, that is, taking the communication characteristics of the regional center as the clustering center, by calculating the similarity between the communication characteristics corresponding to each terminal and the communication characteristics of the regional center, the terminals corresponding to the communication characteristics whose similarity with the communication characteristics of the regional center is greater than a preset threshold are divided into the spatial area where the communication characteristics of the regional center are located, thereby determining the spatial area corresponding to each terminal.

[0091] In this embodiment, the communication characteristics and regional center communication characteristics corresponding to each terminal are determined by analyzing the signal information of each beam level; and space division is performed according to the communication characteristics and regional center communication characteristics corresponding to each terminal. In other words, the communication characteristics corresponding to each terminal can be clustered based on the determined regional center communication characteristics to perform space division, so as to determine the spatial area corresponding to each terminal. This method of performing space division is quick and easy to implement, and can improve the efficiency of the space division method.

[0092] In one embodiment, when determining the communication characteristics corresponding to each terminal and the communication characteristics of the regional center, it can be determined by a pre-trained target deep learning model. The pre-trained target deep learning model includes a target encoding subnetwork and a target query subnetwork, and the output end of the target encoding subnetwork is connected to the input end of the target query subnetwork. Based on this, as shown in FIG4, it involves analyzing the signal information of each beam level to determine an implementation method of the communication characteristics corresponding to each terminal and the communication characteristics of the regional center, and the steps of the implementation method include:

[0093] Step 400: Input each beam-level signal information into the target coding subnetwork in the target deep learning model to obtain the communication characteristics corresponding to each terminal.

[0094] The base station controller inputs the acquired beam-level signal information of each terminal into the target coding subnetwork in the pre-trained target deep learning model. By performing feature extraction through the target coding subnetwork, the communication features corresponding to each terminal can be obtained.

[0095] In an optional embodiment, if the communication feature of the terminal is an angular power spectrum, the dimension of the angular power spectrum is N-dimensional, and the beam-level signal information of each terminal acquired by the base station controller is M-dimensional, where N is greater than M. After the M-dimensional beam-level signal information is input into the target coding subnetwork, the target coding subnetwork maps the M-dimensional beam-level signal information to an N-dimensional vector, that is, the beam-level signal information is dimensionally upgraded to obtain the corresponding communication feature.

[0096] Step 410: Input the communication features corresponding to each terminal into the target query subnetwork in the target deep learning model to obtain the regional center communication features.

[0097] After obtaining the communication characteristics corresponding to each terminal, the base station controller inputs the communication characteristics corresponding to each terminal into the target query subnetwork, and queries the communication characteristics corresponding to each terminal through the target query subnetwork to obtain the communication characteristics of the regional center.

[0098] In this embodiment, by inputting each beam-level signal information into the target coding subnetwork in the target network communication, the communication characteristics corresponding to each terminal can be obtained; by inputting the communication characteristics corresponding to each terminal into the target query subnetwork in the target deep learning model, the regional center communication characteristics can be obtained. This method of determining the communication characteristics corresponding to each terminal and the regional center communication characteristics is quick and easy to implement; and it can improve the accuracy of the communication characteristics corresponding to each terminal and the regional center communication characteristics, thereby improving the accuracy of the spatial area corresponding to each terminal, and further improving the benefits of network optimization and wireless resource management for each spatial area.

[0099] In one embodiment, as shown in FIG5 , another implementation method of analyzing each beam-level signal information to determine the communication characteristics corresponding to each terminal and the communication characteristics of the regional center is provided, and the steps of the implementation method include:

[0100] Step 500: construct a spatial partitioning objective function according to the signal information of each beam level, the communication characteristics corresponding to each terminal, and the communication characteristics of the regional center.

[0101] The base station controller establishes a space partitioning objective function based on the acquired beam-level signal information of each terminal, the determined communication characteristics corresponding to each terminal and the regional center communication characteristics. The space partitioning objective function can ensure that the communication characteristics of each control area after division are consistent.

[0102] In an optional embodiment, it is assumed that the base station controller obtains the beam-level signal information (RSRP) of n terminals, which is expressed as , there exists a mapping subfunction g: , so that ,in, is the communication feature corresponding to the ith terminal. Assume that there are K regional center communication features, that is, the number of divided spatial regions is K, and the K spatial region center communication features , so that ,in, is a metric function to measure the similarity of communication features. The established space partition objective function can be expressed as: .

[0103] The problem of estimating the angular power spectrum can be transformed into a sparse optimization problem using the Localized Statistical Channel Model (LSCM). Assume that the base station is equipped with a uniform antenna array with a total of antenna units. With the (1, 1)th antenna unit as the center, the free space is evenly divided into Pitch angle and Azimuth , we can get The relationship between beam-level RSRP and angular power spectrum can be expressed as:

[0104] Formula (1).

[0105] in, is the RSRP of the mth beam at time t, for The complex channel gain at time t in the direction indicated by the spatial angle, is the antenna transmission coefficient (antenna transmission gain) in the direction indicated by the spatial angle of the m-th beam. Matrixizing the above formula (1) can obtain , where . is the RSRP of M beams, is the desired signal gain in each direction, i.e., the angular power spectrum, is the antenna transmission gain matrix, and A can be expressed as: . Since the number of scatterers in space is limited and the physical environment is stable in a short time, it can be considered that when t < T, the angular power spectrum x remains unchanged and sparse. A is controlled by antenna parameters and can be regarded as known. y can be obtained by averaging the RSRP data collected at a preset position (geographical location unchanged) for a period of time. Estimating the angular power spectrum can be a sparse optimization problem: , and the constraint conditions of this sparse optimization problem are: , where L is the maximum number of signal paths, generally less than 10. Considering the situation of space division, the above sparse optimization problem can be further expressed as: , and the constraint conditions can be expressed as . is the center of the RSRP in the k-th spatial region, is the center of the angular power spectrum in the k-th spatial region, i.e., the regional center communication feature of the k-th spatial region.

[0106] When the communication features corresponding to each terminal obtained are the angular power spectrum, the space division objective function can be expressed as: , and the constraint conditions include , , where is the regularization term coefficient.

[0107] Step 510: Solve the space division objective function to obtain the mapping sub-function and the regional center communication feature.

[0108] After obtaining the space division objective function, the base station controller solves this space division objective function, and can obtain the mapping sub-function g and the regional center communication feature c in the space division objective function.

[0109] Step 520: Determine the communication features corresponding to each terminal according to the mapping sub-function and the signal information at each beam level.

[0110] After obtaining the mapping sub-function, the base station controller can determine the communication features corresponding to each terminal according to this mapping sub-function and the signal information at each beam level. That is, using the mapping sub-function, map the signal at each beam level into the communication features corresponding to each terminal.

[0111] In this embodiment, a spatial division objective function is constructed according to each beam-level signal information, the communication characteristics corresponding to each terminal, and the communication characteristics of the regional center; solving the spatial division objective function can obtain a mapping sub-function and a regional center communication characteristic; according to the mapping sub-function and each beam-level signal information, the communication characteristics corresponding to each terminal are determined. In this way, spatial division is performed according to the regional center communication characteristics obtained by solving the spatial division objective function and the communication characteristics corresponding to each terminal, which can ensure that the communication characteristics of all terminals in each divided spatial area are consistent. In addition, this embodiment provides another method for determining the communication characteristics corresponding to each terminal and the regional center communication characteristics. The user can select the method provided in this embodiment according to the actual application, so that the spatial division method has higher applicability.

[0112] In one embodiment, as shown in FIG6 , the steps of the target deep learning model training method include:

[0113] Step 600: Obtain beam-level signal information samples.

[0114] The beam-level signal information samples may be pre-stored in a storage device of the base station controller. When the target deep learning model needs to be trained, the base station controller obtains the beam-level signal information samples from the corresponding storage device. The description of the beam-level signal information samples can refer to the specific description of the beam-level signal information in the above embodiment, which will not be repeated here.

[0115] Step 610: Input the beam-level signal information samples into the initial deep learning model to obtain reconstructed beam-level signal information.

[0116] The base station controller inputs the acquired beam-level signal information samples into the initial deep learning model, and processes the beam-level signal information samples through the initial deep learning model to obtain reconstructed beam-level signal information. This embodiment does not limit the type and structure of the initial deep learning model, as long as its function can be achieved.

[0117] Step 620: Determine reconstruction loss according to the reconstructed beam-level signal information and the beam-level signal information samples.

[0118] After obtaining the reconstructed beam-level signal information, the base station controller calculates the reconstruction loss according to the reconstructed beam-level signal information and the beam-level signal information sample.

[0119] In an optional embodiment, the reconstruction loss can be measured using the mean square error. Specifically, the reconstruction loss is It can be expressed as: ,in represents the initial deep learning model, Represents the reconstructed beam-level signal information.

[0120] Step 630: Train the initial deep learning model according to the reconstruction loss to obtain a target deep learning model.

[0121] After obtaining the reconstruction loss, the base station controller adjusts the network parameters of the initial deep learning model according to the reconstruction loss to train the initial deep learning model to obtain the target deep learning model.

[0122] In an optional embodiment, the training termination condition for the initial deep learning model may be that the number of training times reaches a preset threshold, or that the reconstruction loss converges.

[0123] In this embodiment, the reconstruction loss is determined according to the reconstructed beam-level signal information and the beam-level signal information samples obtained by inputting the acquired beam-level signal information samples into the initial deep learning model; the initial deep learning model is trained according to the reconstruction loss to obtain the target deep learning model. This method of training the initial deep learning model is fast and easy to implement.

[0124] In one embodiment, the beam-level signal information in the acquired beam-level signal information samples may lack data. Based on this, as shown in FIG7 , an implementation method of inputting the beam-level signal information samples into an initial deep learning model to obtain reconstructed beam-level signal information includes the following steps:

[0125] Step 700: When there is missing data in the beam-level signal information samples, determine a missing vector.

[0126] After obtaining the beam-level signal information sample, the base station controller detects the beam-level signal information sample to determine whether there is missing data in the beam-level signal information sample. If it is determined that there is missing data in the beam-level signal information, the missing vector corresponding to the beam-level signal information sample is determined. Assume that the beam-level signal information sample with missing data is expressed as , The corresponding missing vector s can be expressed as ,in, In the missing data, , Where there is no missing data, .

[0127] Step 710: Complete the missing data in the beam-level signal information samples to obtain completed beam-level signal information samples.

[0128] When determining that there is missing data in the beam-level signal information sample, the base station controller completes the missing data in the beam-level signal information sample to obtain the completed beam-level signal information sample. This embodiment does not limit the specific method of completing the missing data in the beam-level signal information sample, as long as its function can be achieved.

[0129] In an optional embodiment, a preset constant may be used to complete missing data in the beam-level signal information samples. Alternatively, the mean of the non-missing data in the beam-level signal information samples may be calculated and used to complete missing data in the beam-level signal information samples.

[0130] Step 720: Input the completed beam-level signal information samples into the initial deep learning model to obtain reconstructed beam-level signal information.

[0131] After obtaining the completed beam-level signal information samples, the base station controller inputs the completed beam-level signal information samples into the initial deep learning model. The completed beam-level signal information samples are processed by the initial deep learning model to obtain reconstructed beam-level signal information.

[0132] In the case where the acquired beam-level signal information samples have missing data, an implementation method for determining a reconstruction loss according to the reconstructed beam-level signal information and the beam-level signal information samples includes:

[0133] The reconstruction loss is determined according to the reconstructed beam-level signal information, the completed beam-level signal information samples and the missing vector.

[0134] In the case where the acquired beam-level signal information samples have missing data, the base station controller can obtain reconstructed beam-level signal information by inputting the completed beam-level signal information samples into the initial deep learning model. The reconstruction loss can be obtained according to the reconstructed beam-level signal information, the completed beam-level signal information samples and the missing vector.

[0135] In an optional embodiment, when there are missing data in the beam-level signal information samples, when calculating the reconstruction loss, only the reconstruction loss (mask reconstruction loss) where there is no missing data needs to be calculated. It can be expressed as: .

[0136] In this embodiment, when there is missing data in the beam-level signal information sample, the missing vector is determined, and the missing data in the beam-level signal information sample is completed to obtain the completed beam-level signal information sample; the initial deep learning model is trained according to the completed beam-level signal information sample to obtain the target deep learning model. The target deep learning model trained in this way is suitable for the case where there is missing data in the input beam-level signal information sample, which can improve the applicability of the trained target deep learning model, thereby improving the accuracy of the communication characteristics corresponding to each terminal and the communication characteristics of the regional center determined according to the target deep learning model, and then improving the accuracy of the divided spatial area, and improving the benefits of network optimization and wireless resource management for each spatial area.

[0137] In one embodiment, the structure of the initial deep learning model is shown in FIG8 , and the initial deep learning model includes a first initial encoding subnetwork, a first initial query subnetwork, and a first initial decoding subnetwork. The output end of the first initial encoding subnetwork is connected to the input end of the first initial query subnetwork, and the output end of the first initial query subnetwork is connected to the input end of the first initial decoding subnetwork. In this case, as shown in FIG9 , a method for implementing the reconstructed beam-level signal information by inputting a beam-level signal information sample into the initial deep learning model is involved, and the steps of the implementation method include:

[0138] Step 900: Input the beam-level signal information sample into the first initial coding subnetwork to obtain a communication feature sample.

[0139] After obtaining the beam-level signal information sample, the base station controller inputs the beam-level signal information sample into the first initial coding subnetwork, and extracts features of the beam-level signal information sample through the first initial coding subnetwork to obtain communication feature samples corresponding to the beam-level signal information sample.

[0140] In an optional embodiment, the beam-level signal information sample is , if the determined communication feature sample is an angular power spectrum. The first initial encoding subnetwork maps the beam-level signal information sample to a continuous latent variable (i.e., the communication feature sample) , the first initial encoding subnetwork can be expressed as : ,in, is the network weight parameter of the first initial encoding subnetwork.

[0141] Step 910: Input the communication feature sample into the first initial query subnetwork to obtain the regional center communication feature sample.

[0142] The base station controller inputs the communication feature samples outputted from the output end of the first initial encoding subnetwork into the first initial query subnetwork, and queries the regional center communication feature samples from the communication feature samples through the first initial query subnetwork.

[0143] In an optional embodiment, in the space where the latent variables are located, there is a codebook consisting of K discrete latent variables: The first initial query subnetwork can be expressed as : The first initial query subnetwork obtains the discrete latent variable closest to the input latent variable (communication feature sample) in the codebook by searching, which is used as the discrete representation of the latent variable (regional center communication feature sample): .

[0144] Step 920: Input the regional center communication feature sample into the first initial decoding sub-network to obtain reconstructed beam-level signal information; the first initial decoding sub-network includes a neural network or a physical model.

[0145] The base station controller inputs the regional center communication feature samples outputted from the output end of the first initial query subnetwork to the input end of the first initial decoding subnetwork, reconstructs the regional center communication feature samples through the first initial decoding subnetwork, and obtains reconstructed beam-level signal information.

[0146] In an optional embodiment, the region center communication feature samples are mapped back to the input space through the first initial decoding sub-network : , and obtain the reconstructed beam-level signal information ,in, is the network weight parameter of the first initial decoding sub-network.

[0147] The first initial decoding subnetwork may include a multi-layer neural network or a physical model. If the acquired beam-level signal information sample is RSRP data, in order to ensure that the intermediate latent space has the physical meaning of the communication characteristics, the first initial decoding subnetwork may include a channel model in the physical model to represent it. At this time, the reconstructed beam-level signal information can be represented as , the parameters of the first initial decoding subnetwork A is the antenna transmission gain matrix of the base station. This makes the physical model the first initial decoding subnetwork, makes the physical meaning of the latent variable the angular power spectrum, the latent space the angular power spectrum space, and the discrete representation in the codebook is the angular power spectrum in the spatial region.

[0148] In this embodiment, the initial deep learning model includes a first initial encoding subnetwork, a first initial query subnetwork and a first initial decoding subnetwork. By inputting beam-level signal information samples into the first initial encoding subnetwork, a communication feature sample is obtained; the communication feature sample is input into the first initial query subnetwork to obtain a regional center communication feature sample; the regional center communication feature is input into the first initial decoding subnetwork to obtain reconstructed beam-level signal information. The initial deep learning model provided in this embodiment has a simple structure and is easy to train, and can improve the efficiency and accuracy of the target deep learning model obtained by training.

[0149] In one embodiment, as shown in FIG10 , an implementation method of training an initial deep learning model according to a reconstruction loss to obtain a target deep learning model includes the following steps:

[0150] Step 101: Determine discretization loss based on communication feature samples and regional center communication feature samples.

[0151] After obtaining the communication feature samples and the regional center communication feature samples, the base station controller calculates the discretization loss according to the communication feature samples and the regional center communication samples. Optionally, the discretization loss can be expressed as .

[0152] Step 102: train the initial deep learning model according to the reconstruction loss and the discretization loss, and determine the first initial encoding subnetwork and the first initial query subnetwork in the trained initial deep learning model as the target deep learning model.

[0153] After obtaining the discretization loss, the base station controller trains the initial deep learning model according to the determined reconstruction loss and discretization loss, that is, adjusts the network weight parameters of the first initial encoding subnetwork, the network weight parameters of the first initial query subnetwork, and the network weight parameters of the first initial decoding subnetwork in the initial deep learning model. The trained first initial encoding subnetwork is determined as the target encoding subnetwork in the target deep learning model, and the trained first initial query subnetwork is determined as the target query subnetwork in the target deep learning model.

[0154] In this embodiment, the discretization loss is determined by the communication feature samples and the regional center communication feature samples; the initial deep learning model is trained according to the reconstruction loss and the discretization loss, and the first initial encoding subnetwork and the first initial sub-search subnetwork in the trained initial deep learning model are determined as the target deep learning model. In this way, considering the error existing when the regional center communication feature samples are obtained through the first initial query subnetwork and the regional center communication feature samples are input into the first initial decoding subnetwork, the accuracy of the obtained target deep learning model can be improved, thereby improving the accuracy of the communication features corresponding to each terminal and the regional center communication features determined by the target deep learning model, and then improving the accuracy of the divided spatial area, and improving the benefits of network optimization and wireless resource management for each spatial area.

[0155] In an optional embodiment, the expression of the space partitioning objective function can be converted into an expression for model training, which is expressed as: , the constraints include , . The constraints Can be added to the expression of model training, the constraints in The output of the last layer in the first initial decoding subnetwork can be passed through The reconstruction loss can be expressed as: , the discretization loss can be expressed as: Based on this, an implementation method of adjusting the network parameters of the initial deep learning model according to the reconstruction loss and the discretization loss is involved, and the implementation method includes:

[0156] Based on the straight-through estimation method or the soft vector quantization method, the network parameters of the initial deep learning model are adjusted according to the reconstruction loss and the discretization loss.

[0157] The core idea of ​​the straight-through estimator (STE) method is to let the gradient pass directly during the gradient back propagation, by introducing the operator : On this basis, the expression of model training can be converted to:

[0158]

[0159] When forward propagating, the reconstruction loss can be expressed as:

[0160]

[0161] The discretization loss can be expressed as:

[0162]

[0163] When backpropagating the gradient, , , , , ,

[0164] , which is equivalent to The gradient back propagation when directly input to the first initial decoding subnetwork is to let the gradient pass directly without being affected by Operation About Unguided influence.

[0165] Based on the soft vector quantization method, It can be expressed as , the distance between the input latent variable and the discrete latent variable in the codebook is set as the weight, and the discrete latent variable in the codebook is weighted as the discrete representation of the latent variable, so that the gradient can be propagated in a direction through the vector quantization component. Based on this, the expression of model training can be converted to: , constraints can include , .

[0166] In this embodiment, the network parameters of the initial deep learning model are adjusted according to the reconstruction loss and the discretization loss through the direct estimation method or the soft quantization method. This provides two methods for training the initial deep learning model through gradient direction propagation. Users can choose according to actual applications, which can improve the applicability of the space division method.

[0167] In one embodiment, the structure of the initial deep learning model is shown in FIG11, and the initial deep learning model includes a second initial encoding subnetwork, a second initial query subnetwork, and a second initial decoding subnetwork. The output end of the second initial encoding subnetwork is connected to the input end of the second initial query subnetwork, and the output end of the second initial query subnetwork is connected to the input end of the second initial decoding subnetwork. In this case, as shown in FIG12, another implementation method of reconstructing beam-level signal information by inputting beam-level signal information samples into the initial deep learning model is obtained, and the steps of the implementation method include:

[0168] Step 121: Input the beam-level signal information sample into the second initial coding subnetwork to obtain a dimension-reduced vector.

[0169] After obtaining the beam-level signal information sample, the base station controller inputs the beam-level signal information sample into the second initial coding subnetwork, and performs dimensionality reduction processing on the beam-level signal information sample through the second initial coding subnetwork to obtain a reduced dimensionality vector. This embodiment does not limit the dimension of the reduced dimensionality vector after dimensionality reduction processing.

[0170] In an optional embodiment, the beam-level signal information sample is , the second initial coding subnetwork compresses the beam-level signal information samples to obtain a reduced-dimensional vector, and the reduced-dimensional vector j can be expressed as , , D <M, is the network weight parameter of the second initial encoding sub-network.

[0171] In one embodiment, the beam-level signal information sample includes geographic location information of the terminal, and the dimension of the dimension reduction vector is the same as the dimension of the geographic location information.

[0172] The base station controller includes the geographic location information of each terminal in the beam-level signal information of each terminal, and the dimension of the reduced-dimensional vector obtained by using the beam-level signal information sample to input into the second initial coding subnetwork can be the same as the dimension of the geographic location information. Specifically, the beam-level signal information includes the geographic location information , you can add penalty terms during training , let the geographic location be the label of the latent space, and semi-supervise the network parameters of the second initial encoding sub-network.

[0173] In this embodiment, the beam-level signal information sample includes the geographic location information of the terminal, and the dimension of the dimensionality reduction vector obtained by the second initial coding subnetwork is directly set to the dimension of the geographic location information. In this way, semi-supervised training of the network parameters of the second initial coding subnetwork can be achieved, and the accuracy of the target coding subnetwork in the trained target deep learning model can be improved, thereby improving the accuracy of the communication characteristics corresponding to each terminal and the regional center communication characteristics determined by the target deep learning model, thereby improving the accuracy of the divided spatial areas, and improving the benefits of network optimization and wireless resource management for each spatial area.

[0174] Step 122: Input the dimension reduction vector into the second initial query sub-network to obtain a regional center communication feature sample.

[0175] The base station controller inputs the dimension reduction vector outputted from the output terminal of the second initial coding subnetwork into the second initial query subnetwork, determines the dimension reduction regional center communication feature sample through the second initial query subnetwork, and performs dimension increase on the dimension reduction regional center communication feature sample to obtain the regional center communication feature sample. That is, the second initial query subnetwork includes a query module and a dimension increase module, inputs the dimension reduction vector into the query module, determines the dimension reduction regional center communication feature sample through the query module, inputs the dimension reduction regional center communication feature sample into the dimension increase module, performs dimension increase on the dimension reduction regional center communication feature sample through the dimension increase module, and obtains the regional center communication feature sample.

[0176] In an optional embodiment, in the space where the dimension reduction vector is located, there is a codebook consisting of K latent variables The query module can be expressed as : The second initial query subnetwork finds the discrete latent variable closest to the input latent variable (communication feature sample) in the codebook as the discrete representation of the latent variable (the communication feature sample in the center of the reduced dimension): The dimension-raising module can be expressed as : Optionally, the dimension-raising module can be a layer of trainable linear mapping, , W is a linear mapping function. The dimension-raising module can also be a type of pre-set mapping with good distance-preserving properties, that is, if and In potential Specific measurements in space The following is similar, and In potential Specific measurements in space This can ensure the accuracy of the communication feature samples in the center of the region after dimensionality increase, thereby ensuring the consistency of the communication features in each spatial region after division.

[0177] Step 123: Input the regional center communication feature sample into the second initial decoding subnetwork to obtain reconstructed beam-level signal information.

[0178] The base station controller inputs the regional center communication feature sample outputted from the output end of the second initial query subnetwork to the input end of the second initial decoding subnetwork, and reconstructs the regional center communication feature sample through the second initial decoding subnetwork to obtain reconstructed beam-level signal information. The description of inputting the regional center communication feature sample to the second initial decoding subnetwork to obtain the reconstructed beam-level signal information can refer to the specific description of inputting the regional center communication feature sample to the first initial decoding subnetwork to obtain the reconstructed beam-level signal information in the above embodiment, which will not be repeated here.

[0179] In this embodiment, the initial deep learning model includes a second initial encoding subnetwork, a second initial query subnetwork, and a second initial decoding subnetwork. The beam-level signal information sample is input into the second initial encoding subnetwork to obtain a dimension reduction vector; the dimension reduction vector is input into the second initial query subnetwork to obtain a regional center communication feature sample; the regional center communication feature sample is input into the second initial decoding subnetwork to obtain reconstructed beam-level signal information. In this way, the beam-level signal information sample is subjected to dimension reduction processing and then trained, which can improve the learning ability of the initial deep learning model, thereby improving the efficiency of training the initial deep learning model.

[0180] Please refer to FIG. 13 . An embodiment of the present application provides a space division method. The steps of the method include:

[0181] Step 131, obtaining beam-level signal information samples;

[0182] Step 132: input the beam-level signal information sample into the first initial coding subnetwork to obtain a communication feature sample;

[0183] Step 133: input the communication feature sample into the first initial query subnetwork to obtain the regional center communication feature sample;

[0184] Step 134: input the regional center communication feature sample into the first initial decoding subnetwork to obtain reconstructed beam-level signal information;

[0185] Step 135: determining a reconstruction loss according to the reconstructed beam-level signal information and the beam-level signal information sample, and determining a discretization loss according to the communication feature sample and the regional center communication feature sample;

[0186] Step 136: training the network parameters of the initial deep learning model according to the reconstruction loss and the discretization loss, and determining the trained first initial encoding subnetwork as the target encoding subnetwork, and determining the trained first initial query subnetwork as the target query subnetwork;

[0187] Step 137: Obtain beam-level signal information of each terminal;

[0188] Step 138: input each beam-level signal information into the target coding subnetwork to obtain the communication characteristics corresponding to each terminal;

[0189] Step 139: input the communication characteristics corresponding to each terminal into the target query subnetwork to obtain the communication characteristics of the regional center;

[0190] Step 141: perform spatial division according to the communication characteristics corresponding to each terminal and the communication characteristics of the regional center, and determine the spatial area corresponding to each terminal; the communication characteristics corresponding to all terminals in each spatial area are consistent.

[0191] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0192] Based on the same inventive concept, the embodiment of the present application also provides a space division device for implementing the space division method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the one or more space division device embodiments provided below can refer to the limitations on the space division method above, and will not be repeated here.

[0193] In one embodiment, as shown in FIG14 , a space division device 10 is provided, including: an acquisition module 11 and a division module 12, wherein:

[0194] The acquisition module 11 is used to acquire beam-level signal information of each terminal.

[0195] The division module 12 is used to determine the communication characteristics corresponding to each terminal according to the signal information of each beam level, and to perform spatial division according to the communication characteristics corresponding to each terminal to determine the spatial area corresponding to each terminal; the communication characteristics corresponding to all terminals in each spatial area are consistent.

[0196] In one embodiment, the division module 12 includes an analysis unit and a division unit. The analysis unit is used to analyze the signal information of each beam level to determine the communication characteristics corresponding to each terminal and the communication characteristics of the regional center. The division unit is used to perform spatial division according to the communication characteristics corresponding to each terminal and the communication characteristics of the regional center to determine the spatial area corresponding to each terminal.

[0197] In one embodiment, the analysis unit is specifically used to input each beam-level signal information into the target encoding subnetwork in the target deep learning model to obtain the communication characteristics corresponding to each terminal; and input the communication characteristics corresponding to each terminal into the target query subnetwork in the target deep learning model to obtain the regional center communication characteristics.

[0198] In one embodiment, the analysis unit is specifically used to construct a space partitioning objective function based on the signal information of each beam level, the communication characteristics corresponding to each terminal, and the communication characteristics of the regional center; solve the space partitioning objective function to obtain the mapping sub-function and the communication characteristics of the regional center; determine the communication characteristics corresponding to each terminal based on the mapping sub-function and the signal information of each beam level.

[0199] In one embodiment, the space division device 10 further includes a training module. The training module includes an acquisition unit, an input unit, a determination unit, and a training unit. The acquisition unit is used to acquire beam-level signal information samples. The input unit is used to input the beam-level signal information samples into the initial deep learning model to obtain reconstructed beam-level signal information. The determination unit is used to determine the reconstruction loss based on the reconstructed beam-level signal information and the beam-level signal information samples. The training unit is used to train the initial deep learning model according to the reconstruction loss to obtain a target deep learning model.

[0200] In one embodiment, the input unit is specifically used to determine the missing vector when there is missing data in the beam-level signal information sample; to complete the missing data in the beam-level signal information sample to obtain the completed beam-level signal information sample; to input the completed beam-level signal information sample into the initial deep learning model to obtain the reconstructed beam-level signal information; the determination unit is specifically used to determine the reconstruction loss according to the reconstructed beam-level signal information, the completed beam-level signal information sample and the missing vector.

[0201] In one embodiment, the input unit is specifically used to input the beam-level signal information sample into the first initial encoding subnetwork to obtain a communication feature sample; input the communication feature sample into the first initial query subnetwork to obtain a regional center communication feature sample; input the regional center communication feature sample into the first initial decoding subnetwork to obtain reconstructed beam-level signal information; the first initial decoding subnetwork includes a neural network or a physical model.

[0202] In one embodiment, the training unit is specifically used to determine the discretization loss based on the communication feature samples and the regional center communication feature samples; train the initial deep learning model based on the reconstruction loss and the discretization loss, and determine the first initial encoding subnetwork and the first initial query subnetwork in the trained initial deep learning model as the target deep learning model.

[0203] In one embodiment, the training unit is further used to train the initial deep learning model based on a direct estimation method or a soft vector quantization method according to the reconstruction loss and the discretization loss.

[0204] In one embodiment, the input unit is specifically used to input the beam-level signal information sample into the second initial encoding sub-network to obtain a reduced-dimensionality vector; input the reduced-dimensionality vector into the second initial query sub-network to obtain a regional center communication feature sample; input the regional center communication feature sample into the second initial decoding sub-network to obtain reconstructed beam-level signal information.

[0205] In one embodiment, the beam-level signal information sample includes geographic location information of the terminal, and the dimension of the dimension reduction vector is the same as the dimension of the geographic location information.

[0206] Each module in the above-mentioned space division device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0207] In one embodiment, a computer device is provided, and its internal structure diagram may be shown in FIG15. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a space division method is implemented. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, a trackball or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0208] Those skilled in the art will understand that the structure shown in FIG. 15 is merely a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0209] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0210] Obtaining beam-level signal information of each terminal;

[0211] The communication characteristics corresponding to each terminal are determined according to the signal information of each beam level, and space division is performed according to the communication characteristics corresponding to each terminal to determine the space area corresponding to each terminal; the communication characteristics corresponding to all terminals in each space area are consistent.

[0212] In one embodiment, when the processor executes the computer program, the following steps are also implemented: analyzing the signal information of each beam level to determine the communication characteristics corresponding to each terminal and the communication characteristics of the regional center; performing spatial division according to the communication characteristics corresponding to each terminal and the communication characteristics of the regional center to determine the spatial area corresponding to each terminal.

[0213] In one embodiment, when the processor executes the computer program, the following steps are also implemented: inputting each beam-level signal information into the target encoding subnetwork in the target deep learning model to obtain the communication characteristics corresponding to each terminal; inputting the communication characteristics corresponding to each terminal into the target query subnetwork in the target deep learning model to obtain the regional center communication characteristics.

[0214] In one embodiment, when the processor executes the computer program, the following steps are also implemented: constructing a space partitioning objective function according to the signal information of each beam level, the communication characteristics corresponding to each terminal, and the communication characteristics of the regional center; solving the space partitioning objective function to obtain a mapping sub-function and the communication characteristics of the regional center; determining the communication characteristics corresponding to each terminal according to the mapping sub-function and the signal information of each beam level.

[0215] In one embodiment, when the processor executes the computer program, the following steps are also implemented: obtaining beam-level signal information samples; inputting the beam-level signal information samples into the initial deep learning model to obtain reconstructed beam-level signal information; determining the reconstruction loss based on the reconstructed beam-level signal information and the beam-level signal information samples; and training the initial deep learning model based on the reconstruction loss to obtain a target deep learning model.

[0216] In one embodiment, the processor further implements the following steps when executing the computer program: when there is missing data in the beam-level signal information sample, determine the missing vector; complete the missing data in the beam-level signal information sample to obtain the completed beam-level signal information sample; input the completed beam-level signal information sample into the initial deep learning model to obtain reconstructed beam-level signal information; determine the reconstruction loss based on the reconstructed beam-level signal information, the completed beam-level signal information sample and the missing vector.

[0217] In one embodiment, when the processor executes the computer program, the following steps are also implemented: inputting the beam-level signal information sample into the first initial encoding subnetwork to obtain the communication feature sample; inputting the communication feature sample into the first initial query subnetwork to obtain the regional center communication feature sample; inputting the regional center communication feature sample into the first initial decoding subnetwork to obtain the reconstructed beam-level signal information; the first initial decoding subnetwork includes a neural network or a physical model.

[0218] In one embodiment, when the processor executes the computer program, the following steps are also implemented: determining the discretization loss based on the communication feature samples and the regional center communication feature samples; training the initial deep learning model based on the reconstruction loss and the discretization loss, and determining the first initial encoding subnetwork and the first initial query subnetwork in the trained initial deep learning model as the target deep learning model.

[0219] In one embodiment, when the processor executes the computer program, the following steps are also implemented: based on the direct estimation method or the soft vector quantization method, the initial deep learning model is trained according to the reconstruction loss and the discretization loss.

[0220] In one embodiment, when the processor executes the computer program, the following steps are also implemented: inputting the beam-level signal information sample into the second initial encoding sub-network to obtain a reduced-dimensionality vector; inputting the reduced-dimensionality vector into the second initial query sub-network to obtain a regional center communication feature sample; inputting the regional center communication feature sample into the second initial decoding sub-network to obtain reconstructed beam-level signal information.

[0221] In one embodiment, the beam-level signal information sample includes geographic location information of the terminal, and the dimension of the dimension reduction vector is the same as the dimension of the geographic location information.

[0222] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0223] Obtaining beam-level signal information of each terminal;

[0224] The communication characteristics corresponding to each terminal are determined according to the signal information of each beam level, and space division is performed according to the communication characteristics corresponding to each terminal to determine the space area corresponding to each terminal; the communication characteristics corresponding to all terminals in each space area are consistent.

[0225] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: analyzing the signal information of each beam level to determine the communication characteristics corresponding to each terminal and the communication characteristics of the regional center; performing spatial division according to the communication characteristics corresponding to each terminal and the communication characteristics of the regional center to determine the spatial area corresponding to each terminal.

[0226] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented: inputting each beam-level signal information into a target encoding subnetwork in a target deep learning model to obtain a communication feature corresponding to each terminal; inputting the communication feature corresponding to each terminal into a target query subnetwork in a target deep learning model to obtain a regional center communication feature.

[0227] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: constructing a space partitioning objective function based on the signal information of each beam level, the communication characteristics corresponding to each terminal, and the communication characteristics of the regional center; solving the space partitioning objective function to obtain a mapping sub-function and the communication characteristics of the regional center; determining the communication characteristics corresponding to each terminal based on the mapping sub-function and the signal information of each beam level.

[0228] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: obtaining beam-level signal information samples; inputting the beam-level signal information samples into the initial deep learning model to obtain reconstructed beam-level signal information; determining the reconstruction loss based on the reconstructed beam-level signal information and the beam-level signal information samples; training the initial deep learning model based on the reconstruction loss to obtain the target deep learning model.

[0229] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: when there is missing data in the beam-level signal information sample, determining the missing vector; completing the missing data in the beam-level signal information sample to obtain the completed beam-level signal information sample; inputting the completed beam-level signal information sample into the initial deep learning model to obtain reconstructed beam-level signal information; determining the reconstruction loss based on the reconstructed beam-level signal information, the completed beam-level signal information sample and the missing vector.

[0230] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented: inputting beam-level signal information samples into a first initial encoding subnetwork to obtain communication feature samples; inputting communication feature samples into a first initial query subnetwork to obtain regional center communication feature samples; inputting regional center communication feature samples into a first initial decoding subnetwork to obtain reconstructed beam-level signal information; the first initial decoding subnetwork includes a neural network or a physical model.

[0231] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented: determining a discretization loss based on communication feature samples and regional center communication feature samples; training an initial deep learning model based on reconstruction loss and discretization loss, and determining a first initial encoding subnetwork and a first initial query subnetwork in the trained initial deep learning model as target deep learning models.

[0232] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: based on the direct estimation method or the soft vector quantization method, the initial deep learning model is trained according to the reconstruction loss and the discretization loss.

[0233] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: inputting the beam-level signal information sample into the second initial encoding sub-network to obtain a reduced-dimensionality vector; inputting the reduced-dimensionality vector into the second initial query sub-network to obtain a regional center communication feature sample; inputting the regional center communication feature sample into the second initial decoding sub-network to obtain reconstructed beam-level signal information.

[0234] In one embodiment, the beam-level signal information sample includes geographic location information of the terminal, and the dimension of the dimension reduction vector is the same as the dimension of the geographic location information.

[0235] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0236] Obtaining beam-level signal information of each terminal;

[0237] The communication characteristics corresponding to each terminal are determined according to the signal information of each beam level, and space division is performed according to the communication characteristics corresponding to each terminal to determine the space area corresponding to each terminal; the communication characteristics corresponding to all terminals in each space area are consistent.

[0238] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: analyzing the signal information of each beam level to determine the communication characteristics corresponding to each terminal and the communication characteristics of the regional center; performing spatial division according to the communication characteristics corresponding to each terminal and the communication characteristics of the regional center to determine the spatial area corresponding to each terminal.

[0239] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented: inputting each beam-level signal information into a target encoding subnetwork in a target deep learning model to obtain a communication feature corresponding to each terminal; inputting the communication feature corresponding to each terminal into a target query subnetwork in a target deep learning model to obtain a regional center communication feature.

[0240] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: constructing a space partitioning objective function based on the signal information of each beam level, the communication characteristics corresponding to each terminal, and the communication characteristics of the regional center; solving the space partitioning objective function to obtain a mapping sub-function and the communication characteristics of the regional center; determining the communication characteristics corresponding to each terminal based on the mapping sub-function and the signal information of each beam level.

[0241] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: obtaining beam-level signal information samples; inputting the beam-level signal information samples into the initial deep learning model to obtain reconstructed beam-level signal information; determining the reconstruction loss based on the reconstructed beam-level signal information and the beam-level signal information samples; training the initial deep learning model based on the reconstruction loss to obtain the target deep learning model.

[0242] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: when there is missing data in the beam-level signal information sample, determining the missing vector; completing the missing data in the beam-level signal information sample to obtain the completed beam-level signal information sample; inputting the completed beam-level signal information sample into the initial deep learning model to obtain reconstructed beam-level signal information; determining the reconstruction loss based on the reconstructed beam-level signal information, the completed beam-level signal information sample and the missing vector.

[0243] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented: inputting beam-level signal information samples into a first initial encoding subnetwork to obtain communication feature samples; inputting communication feature samples into a first initial query subnetwork to obtain regional center communication feature samples; inputting regional center communication feature samples into a first initial decoding subnetwork to obtain reconstructed beam-level signal information; the first initial decoding subnetwork includes a neural network or a physical model.

[0244] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented: determining a discretization loss based on communication feature samples and regional center communication feature samples; training an initial deep learning model based on reconstruction loss and discretization loss, and determining a first initial encoding subnetwork and a first initial query subnetwork in the trained initial deep learning model as target deep learning models.

[0245] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: based on the direct estimation method or the soft vector quantization method, the initial deep learning model is trained according to the reconstruction loss and the discretization loss.

[0246] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: inputting the beam-level signal information sample into the second initial encoding sub-network to obtain a reduced-dimensionality vector; inputting the reduced-dimensionality vector into the second initial query sub-network to obtain a regional center communication feature sample; inputting the regional center communication feature sample into the second initial decoding sub-network to obtain reconstructed beam-level signal information.

[0247] In one embodiment, the beam-level signal information sample includes geographic location information of the terminal, and the dimension of the dimension reduction vector is the same as the dimension of the geographic location information.

[0248] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0249] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0250] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the present application. It should be noted that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A space division method, characterized in that: The method comprises: Obtaining beam-level signal information of each terminal; The communication characteristics corresponding to each of the terminals are determined according to the beam-level signal information, and space division is performed according to the communication characteristics corresponding to each of the terminals to determine the space area corresponding to each of the terminals; the communication characteristics corresponding to all terminals in each of the space areas are consistent.

2. The method according to claim 1, characterized in that Determining the communication characteristics corresponding to each of the terminals according to the beam-level signal information, and performing spatial division according to the communication characteristics corresponding to each of the terminals to determine the spatial areas corresponding to each of the terminals, includes: Analyze each of the beam-level signal information to determine the communication characteristics corresponding to each of the terminals and the regional center communication characteristics; The space is divided according to the communication characteristics corresponding to each of the terminals and the communication characteristics of the regional center, and the space area corresponding to each of the terminals is determined.

3. The method according to claim 2, characterized in that The analyzing the beam-level signal information to determine the communication characteristics corresponding to each terminal and the regional center communication characteristics includes: Inputting each of the beam-level signal information into the target encoding subnetwork in the target deep learning model to obtain the communication characteristics corresponding to each of the terminals; inputting the communication characteristics corresponding to each of the terminals into the target query subnetwork in the target deep learning model to obtain the regional center communication characteristics; or; A space division objective function is constructed according to the beam-level signal information, the communication characteristics corresponding to each terminal, and the regional center communication characteristics; the space division objective function is solved to obtain a mapping sub-function and the regional center communication characteristics; and the communication characteristics corresponding to each terminal are determined according to the mapping sub-function and the beam-level signal information.

4. The method according to claim 3, characterized in that The training method of the target deep learning model includes: Obtaining beam-level signal information samples; Inputting the beam-level signal information sample into an initial deep learning model to obtain reconstructed beam-level signal information; determining a reconstruction loss according to the reconstructed beam-level signal information and the beam-level signal information sample; Training the initial deep learning model according to the reconstruction loss to obtain the target deep learning model; The step of inputting the beam-level signal information sample into an initial deep learning model to obtain reconstructed beam-level signal information includes: In the case where there is missing data in the beam-level signal information samples, determining a missing vector; Completing the missing data in the beam-level signal information sample to obtain a completed beam-level signal information sample; Inputting the completed beam-level signal information samples into the initial deep learning model to obtain reconstructed beam-level signal information; The determining the reconstruction loss according to the reconstructed beam-level signal information and the beam-level signal information sample comprises: The reconstruction loss is determined according to the reconstructed beam-level signal information, the completed beam-level signal information samples and the missing vector.

5. The method according to claim 4, characterized in that In a case where the initial deep learning model includes a first initial encoding subnetwork, a first initial query subnetwork, and a first initial decoding subnetwork, inputting the beam-level signal information sample into the initial deep learning model to obtain reconstructed beam-level signal information includes: Inputting the beam-level signal information sample into the first initial coding subnetwork to obtain a communication feature sample; Inputting the communication feature sample into the first initial query subnetwork to obtain a regional center communication feature sample; Inputting the regional center communication feature sample into the first initial decoding subnetwork to obtain the reconstructed beam-level signal information; the first initial decoding subnetwork includes a neural network or a physical model; In a case where the initial deep learning model includes a second initial encoding subnetwork, a second initial query subnetwork, and a second initial decoding subnetwork, inputting the beam-level signal information sample into the initial deep learning model to obtain reconstructed beam-level signal information includes: Inputting the beam-level signal information sample into the second initial coding subnetwork to obtain a reduced-dimensionality vector; the beam-level signal information sample includes geographic location information of the terminal, and the dimension of the reduced-dimensionality vector is the same as the dimension of the geographic location information; Inputting the dimension reduction vector into the second initial query sub-network to obtain the regional center communication feature sample; The regional center communication feature sample is input into the second initial decoding subnetwork to obtain the reconstructed beam-level signal information.

6. The method according to claim 5, characterized in that The step of training the initial deep learning model according to the reconstruction loss to obtain the target deep learning model includes: Determining a discretization loss according to the communication feature sample and the regional center communication feature sample; Training the initial deep learning model according to the reconstruction loss and the discretization loss, and determining the first initial encoding subnetwork and the first initial query subnetwork in the trained initial deep learning model as the target deep learning model; The initial deep learning model is trained according to the reconstruction loss and the discretization loss, including: Based on a direct estimation method or a soft vector quantization method, the initial deep learning model is trained according to the reconstruction loss and the discretization loss.

7. A space dividing device, characterized in that: The device comprises: An acquisition module, used to acquire beam-level signal information of each terminal; A division module is used to determine the communication characteristics corresponding to each of the terminals according to the beam-level signal information, and to perform spatial division according to the communication characteristics corresponding to each of the terminals to determine the spatial area corresponding to each of the terminals; the communication characteristics corresponding to all terminals in each of the spatial areas are consistent.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images