Radio map and object distribution map generation method, switching model and training method thereof

Abstract

The invention provides a radio map and object distribution map generation method, a switching model and a training method thereof, and the training method comprises the following steps: generating a non-aligned training set: sampling to generate a scatter radio map based on any actual global radio map in an actual global radio map set; inputting the scattered point radio map and the signal transceiver distribution information into a first generator to obtain an estimated object distribution map; inputting the estimated object distribution map and the signal transceiver distribution information into a second generator to obtain an estimated global radio map; inputting any actual object distribution diagram in the actual object distribution diagram set and the distribution information of the signal transceiver into a second generator to obtain a simulated global radio map; inputting the simulated global radio map and the signal transceiver distribution information into a first generator to obtain a simulated object distribution map; and updating model parameters. According to the method, object distribution and global accurate estimation are realized by using scattered RSRP signals.

Description

Technical Field

[0001] The present invention belongs to the technical field of environment-aware communication, and in particular relates to a method for generating a radio map and an object distribution map, a switching model and a training method thereof. Background Art

[0002] In the rapidly developing digital age, communication systems need to have higher perception and adaptability to support efficient and stable communications, especially in the context of smart manufacturing and industrial Internet. Environmentally aware communication technology acquires and analyzes environmental information, optimizes communication performance, improves transmission efficiency, and achieves accurate environmental monitoring, thereby promoting the intelligentization of the industry.

[0003] Although existing technologies have made some progress in environmentally aware communications, they still face significant technical challenges and deficiencies in practical applications. Among them, the estimation of urban building distribution and the estimation of indoor personnel and obstacle distribution are two important technical issues. Current wireless propagation models often rely on dense signal measurement data for global estimation, which not only increases the complexity and cost of data collection, but also cannot accurately reflect the actual distribution of buildings and objects when the data is incomplete or the environment is complex.

[0004] Specifically, when estimating building distribution in urban environments, existing technologies often have difficulty dealing with the complexity and variability of urban structures, resulting in the failure to effectively eliminate signal interference between buildings, affecting the accuracy of the estimation. At the same time, in indoor environments, the dynamic changes of people and obstacles make it impossible for traditional signal measurement methods to reflect their distribution in real time, further restricting the effectiveness of environmental perception communication. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the above-mentioned background technology, and to provide a method for generating a radio map and an object distribution map, a switching model and a training method thereof, so as to achieve accurate estimation of object distribution and global distribution by using only scattered RSRP signals. This innovative method can not only effectively reduce the complexity of data collection, but also infer global signals and distribution information through intelligent algorithms when data is incomplete, thereby improving the performance and adaptability of environmental perception communication. The successful implementation of this technology will bring significant technological breakthroughs to the fields of intelligent manufacturing and environmental monitoring.

[0006] The technical solution adopted by the present invention is: a training method for a switching model of a radio map and an object distribution map, comprising the following steps:

[0007] Generate non-aligned training sets: construct training data containing a set of actual object distribution maps and a set of actual global radio maps;

[0008] Randomly generate signal transceiver device distribution information: generate random signal transceiver device distribution information for each training round of the first generator and the second generator;

[0009] The training process is repeated until the model reaches the desired accuracy:

[0010] Based on any actual global radio map in the actual global radio map set, a scattered radio map is sampled and generated; the scattered radio map and the signal transceiver device distribution information are input into the first generator to obtain an estimated object distribution map; the estimated object distribution map and the signal transceiver device distribution information are input into the second generator to obtain an estimated global radio map; any actual object distribution map in the actual object distribution map set and the signal transceiver device distribution information are input into the second generator to obtain a simulated global radio map; the simulated global radio map and the signal transceiver device distribution information are input into the first generator to obtain a simulated object distribution map;

[0011] Calculate loss functions and update model parameters: calculate the consistency loss function between the estimated global radio map and the actual global radio map; calculate the generative adversarial loss function between the simulated global radio map and the actual global radio map; calculate the consistency loss function between the simulated object distribution map and the actual object distribution map; calculate the generative adversarial loss function between the estimated object distribution map and the actual object distribution map; based on the above loss functions, update the parameters of the first generator and the second generator.

[0012] In the above technical solution, during a single round of training of the first generator and the second generator, the distribution information of the signal transceiver devices used is the same.

[0013] In the above technical solution, the area types corresponding to the actual object distribution map set and the actual global radio map set are the same as the area types of the designated areas; the area types include urban outdoor areas and indoor areas.

[0014] In the above technical solution, the consistency loss function between the estimated global radio map and the actual global radio map uses a mean square error loss function.

[0015] In the above technical solution, the consistency loss function between the simulated object distribution map and the actual object distribution map adopts the dice loss function.

[0016] The present invention provides a switching model of a radio map and an object distribution map, wherein the model is obtained by training using the training method described in the above technical solution, and comprises:

[0017] A first generator: used for generating an object distribution map based on the scattered radio map and the signal transceiver device distribution information;

[0018] The second generator is used to generate a global radio map based on the object distribution map and the signal transceiver device distribution information.

[0019] The present invention provides a method for generating a radio map, which inputs an object distribution map and signal transceiver equipment distribution information of a specified area, or a scattered radio map, into a switching model of the radio map and the object distribution map described in the above technical solution to obtain a global radio map of the specified area.

[0020] The present invention provides a method for generating an object distribution map, which inputs a scattered radio map of a specified area into the switching model of the radio map and the object distribution map described in the above technical solution to obtain the object distribution map of the specified area.

[0021] The present invention provides a computer storage medium, a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method described in the above technical solution is implemented.

[0022] The present invention provides an electronic device, which includes: a processor for executing a computer program stored on the device; and a memory for storing the computer program; wherein, when the computer program is executed by the processor, the method described in the above technical solution is implemented.

[0023] The beneficial effects of the present invention are as follows: the present invention provides a training method for a switching model of a radio map and an object distribution map, which realizes efficient data utilization by generating a non-aligned training set and a cyclic training process. Especially in domain conversion tasks, there is a domain alignment problem, and it is often difficult to achieve pixel-by-pixel alignment between the set domain in the environmental domain and the target set. Since environmental information such as buildings and altitudes are usually provided by satellites, and radio information such as RSRP is collected by road test vehicles or on-site signal receiving terminals, the difficulty in alignment between the two significantly affects the training effect of the model. Therefore, the present invention particularly establishes a non-aligned RSRP data set, fully utilizes the complementarity of environmental information and radio information, optimizes the training of the model, enhances the adaptability and accuracy in complex environments, and provides a reliable data basis for subsequent global estimation.

[0024] Furthermore, the present invention uses the same signal transceiver device distribution information during the training process, which can ensure the input consistency between the first generator and the second generator, reduce the model instability caused by input changes, and thus improve the training efficiency and model performance.

[0025] Furthermore, the present invention ensures that the actual object distribution map and the actual global radio map set are of the same area type as the designated area, which enhances the applicability and accuracy of the model. This feature enables the model to remain effective in different types of environments and adapt to the complexity of urban outdoor and indoor areas.

[0026] Furthermore, the present invention uses the mean square error loss function as the consistency loss function, which can effectively evaluate the error between the estimated global radio map and the actual map, optimize the model parameters, and improve the prediction accuracy of the model. This method is widely used in signal estimation tasks and has good performance.

[0027] Furthermore, the present invention adopts the dice loss function to calculate the consistency loss between the simulated object distribution map and the actual object distribution map, which can better handle the imbalance and sparsity of the data and improve the accuracy of the model in object distribution estimation, especially in complex scenarios.

[0028] Furthermore, the present invention can effectively improve the quality and robustness of the generated model through the specific calculation of the generative adversarial loss function, enhance the adaptability of the model in diverse environments, and ensure that the generated radio map and object distribution map are more realistic and reliable.

[0029] Furthermore, the present invention can flexibly respond to different input data through the switching model of the radio map and the object distribution map obtained through training, quickly generate the corresponding object distribution map and the global radio map, and improve the practicality and efficiency of the system.

[0030] Furthermore, the present invention inputs the object distribution map or scattered radio map of the specified area into the model to generate a high-precision global radio map. This method improves the cognitive ability of the regional radio environment and provides important support for subsequent communication optimization and environmental perception.

[0031] Furthermore, the present invention generates an object distribution map by inputting a scattered radio map of a specified area, which can achieve accurate environmental reconstruction, help improve the efficiency and accuracy of environmental perception communications, and provide data support for smart manufacturing and urban management. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a principle diagram of the training process of the present invention;

[0033] Figure 2 A schematic diagram of radio map domain conversion according to the present invention;

[0034] Figure 3 It is a schematic diagram of a non-aligned data set of the present invention;

[0035] Figure 4 FIG. a is a comparison diagram of an embodiment of the present invention and the prior art;

[0036] Figure 5 FIG b is a comparison diagram of an embodiment of the present invention and the prior art;

[0037] Figure 6Figure c is a comparison diagram between an embodiment of the present invention and the prior art;

[0038] Figure 7 FIG. d is a comparison diagram between an embodiment of the present invention and the prior art;

[0039] Figure 8 The effect diagram a of the embodiment of the present invention is used;

[0040] Fig. 9 The effect diagram b is used for the embodiment of the present invention. DETAILED DESCRIPTION

[0041] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments to facilitate a clear understanding of the present invention, but they do not constitute a limitation on the present invention.

[0042] Example 1

[0043] like Figure 1 As shown, the present invention provides a training method for a switching model of a radio map and an object distribution map, comprising the following steps:

[0044] Generate non-aligned training sets: construct training data containing a set of actual object distribution maps and a set of actual global radio maps;

[0045] Randomly generate signal transceiver device distribution information: generate random signal transceiver device distribution information for each training round of the first generator and the second generator;

[0046] The training process is repeated until the model reaches the desired accuracy:

[0047] Based on any actual global radio map x in the actual global radio map set, a scattered radio map x' is generated by sampling; the scattered radio map x' is combined with the signal transceiver device distribution information (in the outdoor scenario, Figure 1 The base station information shown in (i) is input to the first generator F to obtain an estimated object distribution map F(x'); the estimated object distribution map F(x') and the signal transceiver device distribution information are input to the second generator G to obtain an estimated global radio map G(F(x'));

[0048] Input any actual object distribution map y in the actual object distribution map set and the signal transceiver device distribution information into the second generator G to obtain a simulated global radio map G(y); input the simulated global radio map G(y) and the signal transceiver device distribution information into the first generator F to obtain a simulated object distribution map F(G(y));

[0049] Calculate the loss function and update the model parameters:

[0050] Calculate the consistency loss function between the estimated global radio map G(F(x')) and the actual global radio map x;

[0051] Calculate the generative adversarial loss function between the simulated global radio map G(y) and the actual global radio map x;

[0052] Calculate the consistency loss function between the simulated object distribution map F(G(y)) and the actual object distribution map y;

[0053] Calculate the generative adversarial loss function of the estimated object distribution map F(x') and the actual object distribution map y; based on the above loss function, update the parameters of the first generator and the second generator.

[0054] Specifically, the signal transceiver device distribution information includes: within a specified area (a specific area when the model is applied, or an area randomly generated during model training), the coordinates of all signal transceiver devices in the gridded area and the corresponding received signal power RSRP under their coordinates.

[0055] Specifically, the global radio map is an overall distribution of the radio signal received strength (RSRP) at each coordinate location within a specified area (especially within a city) with a set accuracy. This map provides a comprehensive view of the radio signal coverage, strength changes and attenuation in the area.

[0056] In outdoor scenarios (inside cities), the global radio map shows the comprehensive effect of base station signal coverage in urban areas, including the impact of natural and man-made obstacles in urban environments such as buildings and transportation facilities on radio signal propagation. By integrating multiple measurement data, a spatial distribution map of signal strength is formed to assist in the planning and optimization of communication networks.

[0057] In indoor environments, the global radio map reflects the radio signal strength in various areas of the room, usually generated by signal data collected by fixed sensors or mobile devices at different locations. This map helps identify signal blind spots and interference areas, helping to improve indoor communication quality.

[0058] Specifically, a scattered radio map is a discrete representation of the received signal strength (RSRP) data collected by measuring points with specific known coordinates in a specific area. During the model training process, a scattered radio map is sampled and generated through the actual global radio map.

[0059] Specifically, an object distribution map is a representation of the location information of objects such as buildings, people, and obstacles in a specified area. This diagram is crucial to understanding the impact of the environment on radio signal propagation.

[0060] In outdoor scenes (inside cities), the object distribution map shows the specific location, shape, and height of buildings, helping to analyze the signal propagation path and signal attenuation in urban areas. This diagram is of great significance for the optimization and design of network coverage.

[0061] In indoor environments, object distribution maps show the distribution of people, furniture, and other obstacles. This information can help evaluate the propagation characteristics and quality of indoor signals and ensure the effectiveness of communication systems in complex environments.

[0062] Specifically, the area type corresponding to the actual object distribution map set and the actual global radio map set is the same as the area type of the designated area; the area type includes an urban outdoor area and an indoor area. That is, for indoor application scenarios, the data of the indoor scene is used to construct a corresponding training set and train the switching model. Similarly, for outdoor application scenarios, the data of the outdoor scene is used to construct a corresponding training set and train the switching model.

[0063] Indoor and outdoor scenarios differ significantly in environmental characteristics, signal propagation properties, building layouts, etc. Therefore, using a training set specifically built for a specific scenario (such as indoors or outdoors) ensures that the model is better adapted to these specific environmental conditions, thereby improving the accuracy and reliability of the prediction.

[0064] When constructing the training set, ensuring that the actual object distribution map and global radio map used are the same as the target area type (outdoor or indoor) can ensure data consistency. This helps reduce errors and inconsistencies caused by different data sources and enables the model to learn using similar environmental features during training and inference.

[0065] By training in a specific scenario, the model can focus on extracting important features related to that scenario. For example, in indoor scenarios, the model can learn the signal attenuation characteristics related to furniture, walls, and personnel distribution, while in outdoor scenarios, it can focus on the impact of building height, spacing, terrain, etc. on signal propagation.

[0066] Using scenario-specific data to build training sets can reduce the complexity of the model training process. The similarity of training set features and data structures between different scenarios can simplify the design of the model architecture and reduce unnecessary complexity, thereby accelerating the training process and improving training efficiency.

[0067] When a model is trained in a specific scenario, it can better learn the characteristics of that scenario. This targeted training can improve the model's generalization ability in similar scenarios, enabling it to better cope with unseen samples in practical applications.

[0068] In practical applications, the communication requirements and challenges of indoor and outdoor environments are different. Therefore, building a training set for specific scenarios and training the corresponding switching model can better meet the needs of practical applications and improve the system performance and user experience.

[0069] like Figure 2 As shown, from the perspective of domain transfer, the radio map prediction task is a process of converting from the environmental domain to the radio domain. Conversely, the urban map reconstruction task is an inversion process. Therefore, the present invention considers the global radio map reconstruction task and the object distribution map reconstruction task as dual tasks. Based on the measured RSRP subset X, an end-to-end deep learning model is established to simultaneously reconstruct the building map in the urban area and the radio map of the entire community. The global radio map reconstruction task and the object distribution map reconstruction task are respectively completed by two generators based on generative AI.

[0070] In the domain conversion task, there is a domain alignment problem. The domain alignment problem is that it is often difficult to achieve pixel-by-pixel alignment between the set Z in the environment domain and the target set Y in the domain.

[0071] In indoor scenarios, the reasons why the environmental domain and the radio domain are difficult to align are mainly the following:

[0072] Environmental information collection: Indoor environmental information (such as buildings, walls, furniture, etc.) is usually obtained through 3D modeling, laser scanning or image processing, etc. These methods focus on the geometric shape and distribution characteristics of the physical space.

[0073] Radio information collection: Radio signals (such as RSRP) rely on measurements of radio receivers, usually collected through road test equipment or fixed sensors. The propagation and reception of these signals are affected by many factors, including building materials, signal attenuation and reflection.

[0074] Signal propagation: The propagation characteristics of radio signals in indoor environments are complex and are greatly affected by obstacles (such as walls and furniture), resulting in very uneven distribution of signal strength in space. In contrast, environmental models are usually static and do not consider the dynamic changes of signals during transmission.

[0075] Multipath effect: In indoor environments, radio signals experience reflection, refraction and diffraction, forming multiple signal paths. This multipath effect causes the signal strength at the same location to vary significantly at different time points, increasing the difficulty of alignment.

[0076] Inconsistent data formats: Environmental data and radio data are often stored and represented in different formats. Environmental data may be in the form of 3D models, while radio data is time series or scattered data. This difference complicates direct alignment.

[0077] Lack of annotation information: In environmental data, the locations of buildings and obstacles are clearly defined, while the measurement points of radio signals are often randomly distributed without clear correspondence, making it difficult to achieve precise alignment.

[0078] Differences in data collection time: Environmental data and radio signal data may be collected in different time periods. The indoor environment may change due to renovations, equipment movement, etc., resulting in temporal inconsistencies between the two, further increasing the difficulty of alignment.

[0079] Changes in human flow and equipment: In indoor scenes, changes in human flow, furniture positions, and equipment will affect the propagation characteristics of radio signals. However, environmental models are generally static and cannot reflect these dynamic changes, resulting in an unstable correspondence between the environmental domain and the radio domain.

[0080] In outdoor scenarios, the reasons why the environmental domain and the radio domain are difficult to align are mainly due to the following aspects:

[0081] Environmental information collection: Outdoor environmental information is usually obtained through satellite images, drone photography, laser scanning, or ground measurement. The geographic and building data provided by these methods may be delayed and not updated in a timely manner.

[0082] Radio information collection: Data of radio signals (such as RSRP) are usually collected by mobile measurement vehicles, fixed base stations or portable devices. These signal measurement points may not be consistent with the collection time and location of environmental information, making it difficult to establish a corresponding relationship.

[0083] Signal propagation characteristics: In outdoor environments, the propagation of radio signals is affected by many factors, such as buildings, trees, and terrain. Signals may form complex propagation paths due to reflection, refraction, and diffraction, making the signal strength highly uneven in space.

[0084] Dynamic environmental changes: The outdoor environment is often affected by weather, seasonal changes, and surrounding objects (such as vehicles and pedestrians). These factors can cause the strength and quality of radio signals to change at any time, making the alignment of the environmental domain and the radio domain more difficult.

[0085] Signal interference: In complex scenarios such as cities, signals are often interfered by buildings and other obstacles, forming a multipath effect. This situation makes the radio signal strength measured at different times at the same location significantly different, making the alignment of the radio domain and the environmental domain complicated.

[0086] Inconsistent data formats: Environmental data are usually presented in Geographic Information System (GIS) format, describing static information such as terrain and building locations, while radio data may be stored in time series or scattered point form, lacking clear spatial correspondence.

[0087] Lack of consistent annotation: There may be a lack of unified annotation standards between features in the environment domain (e.g., buildings, roads) and features in the radio domain (e.g., signal strength, coverage), making it difficult to directly align them when building models.

[0088] Data collection time differences: There may be differences in the collection time of environmental data and radio signal data. For example, environmental data may reflect the situation at a specific point in time, while radio signals may be collected at another time period, resulting in poor time consistency between the two.

[0089] Complexity of urban environments: In urban environments, the height, shape, and spacing of buildings have a significant impact on signal propagation. This complexity makes it difficult to accurately match the environmental domain and the radio domain when modeling.

[0090] Temporary changes: Temporary factors such as traffic flow, construction activities, and weather changes can also affect the propagation characteristics of radio signals. This dynamic change increases the difficulty of aligning the environmental domain and the radio domain.

[0091] Considering the above situation, the present invention establishes a non-aligned RSRP dataset (i.e., radio set) and an object distribution dataset (i.e., environment set), such as Figure 3 As shown, the environment set refers to a set of actual object distribution maps, and the radio set refers to a set of actual global radio maps.

[0092] In indoor scenarios, the establishment of the actual global radio map set includes:

[0093] Equipment selection: Choose an appropriate radio measurement device, such as a mobile signal strength meter or a smartphone.

[0094] Survey plan: Plan the survey route, including different floors and rooms, to ensure coverage of the entire building.

[0095] Signal measurement: Collect RSRP data at different locations and time points, making sure to record the coordinates and timestamp of each measurement point. Due to the complex indoor environment, multiple measurements may be required to capture changes in the signal.

[0096] Data storage: The collected RSRP data and the corresponding spatial coordinates (such as room number, coordinate position) are stored in the database.

[0097] In indoor scenes, the establishment of the actual object distribution map set includes:

[0098] Environmental modeling: Generate an indoor environmental model using laser scanning, photo measurement or 3D modeling software. Record information such as the layout of the building and the location of furniture.

[0099] Data Annotation: Annotate the environment model to identify and mark the location and size of buildings, walls, furniture, and obstacles.

[0100] Data integration: Object distribution information is combined with RSRP data to form a comprehensive dataset, but there may be non-alignment due to different measurement time points or methods.

[0101] In outdoor scenarios, the establishment of an actual global radio map set includes:

[0102] Measurement equipment: Use mobile measurement vehicles, drones, or fixed base stations to measure RSRP.

[0103] Measurement routes: Plan measurement routes that cover a specified area (e.g. city streets, parks, etc.), ensuring multiple measurement points.

[0104] Data collection: Take measurements at different time periods (such as peak and off-peak periods) to capture changes in the signal. Record the geographic coordinates and time information of each measurement point.

[0105] Data storage: Store RSRP data with corresponding geographic coordinates in a database, ensuring that signal strength and other relevant information is recorded.

[0106] In indoor scenes, the establishment of the actual object distribution map set includes:

[0107] Satellite imagery: Use images taken by satellites or drones to obtain distribution information of buildings, roads, trees, etc.

[0108] Geographic Information System (GIS): Use GIS software to process and analyze data and build geographic models of the urban environment.

[0109] Data labeling: Label the environmental model to identify and record the specific location and shape of buildings, roads, green spaces, etc.

[0110] Data integration: Combine object distribution data with RSRP data to create a comprehensive dataset, noting that there may be non-alignment due to measurements made at different times and methods.

[0111] Preferably, the processing of the non-aligned actual object distribution map set and the actual global radio map set comprises:

[0112] Differences in time and space: When generating the dataset, ensure that there are certain differences in the collection time and method of RSRP data and object distribution data, which naturally leads to a non-aligned state.

[0113] Differences in collection objects: By collecting building distribution information, indoor distribution information and RSRP data in different areas, a set of actual object distribution maps and a set of actual global radio maps are formed, and there is no need for a one-to-one correspondence between the two. The building data in the actual object distribution map set can be building maps of various shapes, and the RSRP data in the actual global radio map set can be collected in different urban areas or indoor scenes.

[0114] Specifically, during a single round of training of the first generator and the second generator, the distribution information of the signal transceiver devices used is the same.

[0115] Preferably, in each round of training sample generation process, the scheme for automatically and randomly generating the signal transceiver device distribution information comprises the following steps:

[0116] Equipment distribution model: First, a distribution model of signal transceiver equipment is established based on the real environment or simulated environment. The model can take into account factors such as building location, signal attenuation characteristics, and terrain.

[0117] Randomization algorithm: A randomization algorithm is designed to automatically generate a set of distribution information of signal transceiver devices at the beginning of each training round. This information can be randomly laid out based on certain rules (such as device spacing, coverage, etc.) to ensure that the generated layout is within a reasonable range.

[0118] Parameterized configuration: Set the parameters of the device (such as transmit power, frequency, etc.) to ensure that the generated distribution information has consistency in signal characteristics. Parameterization can help more accurately simulate the device layout in the actual environment.

[0119] Data generation: Use the generated signal transceiver device distribution information, combined with the actual or simulated radio environment, to generate corresponding training samples. These samples are used for training the first generator and the second generator.

[0120] Feedback mechanism: After each training round, the generated distribution information and model performance are recorded, and the randomization algorithm and distribution model are adjusted through the feedback mechanism to optimize the subsequent device layout generation process.

[0121] The present invention uses an unsupervised dual estimation method, the framework of which is as follows Figure 1As shown. The input X of the model represents a map drawn based on the RSRP measured at scattered points in the region, called the measured RSRP map, that is, the scattered radio map. The system is equipped with two CNNs-based generators G(·) and F(·), which represent the change and inverse transformation from the environmental domain to the radio domain, respectively. The generator F(·) is used to learn the mapping between the measured RSRP map and the object distribution map, F(X)→Z. Similarly, the generator G is used to learn the mapping, G(Z)→Y. Y represents the global radio map. Z represents the object distribution map.

[0122] The backbone path is marked by black directed lines, and the loss function calculations are green and blue, respectively. The backbone path uses generators F(·) and G(·) to independently complete the conversion between the radio domain and the environment domain. In the backbone path, the input X of generator F(·) and the input Z of generator G(·) predict the object distribution map F(X) and the radio map G(Z) along the black lines, respectively. The goal of the backbone path is to complete the radio map reconstruction and object distribution map reconstruction tasks using the mappings F(X)→Z and G(Z)→Y, respectively.

[0123] In the environmental electromagnetic map reconstruction algorithm, there are two loss functions: generative adversarial loss and consistency loss. Generative adversarial loss is used to verify the similarity between the graph generated by the model and the real graph distribution; consistency loss is used to verify whether the result generated by the model is consistent with the input.

[0124] Generator F needs to convert the radio domain to the environment domain, while the goal of generator G is to convert the environment domain to the radio domain. Since generators F and G complete two different domain conversion tasks, different loss functions are designed for the two tasks.

[0125] On the one hand, the generator G predicts the radio map in the city. Since RSRP is a continuous value, the prediction of the radio map is considered as a regression fit for each pixel in the building map. Therefore, for the radio map reconstruction task applying the generator G, the commonly used mean square error loss function is adopted. Therefore, the consistency loss in the radio domain uses the mean square error loss function, which can be expressed as:

[0126]

[0127] Among them, p i and g i The corresponding pixel prediction values ​​and true values ​​are respectively, and N represents the total number of pixels. During training, the present invention (RadioYnet) automatically adjusts the parameters of the model through a gradient descent algorithm to try to reduce the mean square error between the predicted radio map and the true radio map.

[0128] On the other hand, generator F is used to predict the distribution of buildings in the city or the distribution of indoor personnel obstacles. The prediction of object distribution is considered as a pixel-by-pixel classification of the measured RSRP map. Because the changing trend of the partial derivative of the mean square error loss function is inconsistent with the changing trend of the difference between the predicted value and the true value. Therefore, for the building reconstruction task applying generator F, the mean square error cannot be directly used as the loss function. In the binary classification problem, the binary cross entropy loss function (Binary CrossEntropy Loss, BCE) is a distribution-based loss function. The gradient value calculated by the BCE loss function is proportional to the difference between the reconstructed value and the true value. BCE loss function L BCE It can be expressed as:

[0129]

[0130] Among them, p i and g i The corresponding predicted value and true value of the i-th pixel respectively, and N represents the total number of pixels.

[0131] Different from the distribution-based BCE loss function, the region-based Dice Loss function is a loss function that calculates the similarity of a set, which is used to calculate the similarity of two sample points. D can be expressed as:

[0132]

[0133] Each pixel in the building map has only 0 or 1, corresponding to whether there is a building in the area. From the perspective of the set, the dice coefficient is a measure of the overlap between two sets. The dice coefficient ranges from 0 to 1. The larger the value, the higher the overlap between the predicted building and the actual building distribution. Therefore, the dice loss function is used in the object distribution map reconstruction task to maximize the overlap between the object distribution reconstruction map and the real map. The consistency loss of the environmental domain uses the dice loss function, which can be expressed as:

[0134]

[0135] For the generative adversarial loss, since the input and output of generators G and F share a lot of information, each layer contains all the information, which causes the neural network to be redundant and error-prone. Therefore, the present invention uses a symmetric encoder-decoder structure U-Net to reduce the load of each layer. The downsampling layer is connected to the upsampling layer to adjust the output to the same size because they can be regarded as carrying similar information. For the discriminator D, the features of the input image are extracted through a three-layer downsampling layer structure, and the output is the "true / false" classification of the input image.

[0136] The downsampling layer consists of a convolutional layer, a BN layer, and a leaky rectified linear unit (Leaky ReLU) as an activation function. The upsampling layer consists of a deconvolutional layer, a normalization layer, and a LeakyReLU activation function. The purpose of the convolution operation is to extract different features of the input, and multiple convolutional layers can iteratively extract more complex features. The normalization layer is used to improve the stability of the input distribution of each layer and reduce the interval covariate shift. The generated G′(X) must be similar to the real image y. The optimization objective can be expressed as:

[0137] L GAN (A, B) = Σ x，y logA(x,y)+Σ x log B(x, A(x));

[0138] Among them, A represents the generator, that is, Figure 1 The generator G or F in the above example. B is the discriminator, which uses a multi-layer convolutional neural network structure. x and y are the input and output of the generator respectively. A(x) is the result predicted by the generator A.

[0139] The generator G tries to minimize the loss function to generate the most realistic images, while the adversarial discriminator D tries to maximize the loss function to distinguish pseudo-generated images, which can be expressed as:

[0140]

[0141] Among them, T* is the optimization objective function, the generator A tries to minimize the loss function to generate the most realistic images, while the discriminator B tries to maximize the loss function to distinguish pseudo-generated images.

[0142] Example 2

[0143] The present invention provides a switching model of a radio map and an object distribution map, wherein the model is obtained by training using the training method described in the above technical solution, and comprises:

[0144] A first generator: used for generating an object distribution map based on a scattered radio map;

[0145] The second generator is used to generate a global radio map based on the object distribution map and the signal transceiver device distribution information.

[0146] In the application process of the switching model, the distribution information of the signal transceiver equipment is converted to generate a scattered radio map. That is, the scattered radio map is generated by the RSRP signals of the signal transceiver equipment at several known coordinates in the specified area. The above RSRP signal data is usually obtained by the road test vehicle or the signal receiving equipment at a specific location, reflecting the spatial distribution characteristics of the signal strength.

[0147] In outdoor scenarios (inside cities), the scattered radio map contains multiple measurement points, each of which records the radio signal strength at that location. The map can reveal blank areas of urban signal coverage and potential sources of interference, providing data support for base station layout and network optimization.

[0148] In indoor environments, scattered radio maps are constructed using data points collected from signal receiving devices at different locations. This map helps assess the uniformity and quality of indoor signals and provides an intuitive understanding of the indoor communication environment.

[0149] Preferably, in an outdoor scenario (inside a city), the RSRP signal in the scattered radio map is obtained by:

[0150] Road test vehicle: Use a road test vehicle equipped with a signal receiver and measurement equipment to cruise around the city. The vehicle stops at different locations and records the radio signal strength (RSRP) at each location. This method can cover a large area and provide real-time signal data.

[0151] Mobile devices: Use a smartphone or other mobile device equipped with a radio receiving module to conduct walking tests inside the city. The user collects signal strength data while walking along a designated route or a random route. This method is highly flexible and can collect signal data in crowded or special environments.

[0152] Fixed sensor network: Fixed signal receivers are deployed in urban areas to collect signal strength data regularly. This method is suitable for scenarios that require long-term monitoring and can provide stable signal data collection.

[0153] Drone measurement: Use drones with signal receiving equipment to collect data in complex or difficult-to-reach urban areas. This method can quickly cover a specific area and obtain high-precision signal data.

[0154] Preferably, in an indoor scenario, the RSRP signal in the scattered radio map is obtained by:

[0155] Fixed signal receivers: Multiple fixed signal receivers are placed in the indoor environment to regularly record the signal strength at each location. This method is suitable for offices, shopping malls and other environments and can provide stable signal strength data.

[0156] Mobile device testing: Testing is performed indoors using a mobile device with signal measurement capabilities, such as a smartphone or tablet. The user moves around the room at different locations and records the signal strength, especially in high-traffic areas or near specific obstacles.

[0157] Indoor positioning system: Combined with indoor positioning technology, use devices with known locations to record signal strength. Construct an indoor scattered radio map based on known locations and corresponding signal strengths.

[0158] 3D modeling technology: Use laser scanning or 3D modeling technology in indoor environments, combined with radio signal strength data, to build a 3D model of the indoor environment and extract signal strength at different locations.

[0159] The collected scattered radio signal data are sorted and processed to remove noise and outliers to ensure data quality. Interpolation algorithms (such as Kriging interpolation) are then used to convert the scattered data into a continuous signal strength distribution map as a scattered radio map for subsequent analysis.

[0160] Example 3

[0161] The present invention provides a method for generating a global radio map, which inputs an object distribution map and signal transceiver equipment distribution information of a specified area, or a scattered radio map, into the switching model of the radio map and the object distribution map described in Example 2 to obtain a global radio map of the specified area.

[0162] Specifically, obtaining a global radio map through a scattered radio map is suitable for comprehensive evaluation of signal coverage, especially in large areas (such as cities and indoor environments), and can provide strong support for network optimization, personnel positioning and environmental management. Obtaining a global radio map through an object distribution map can provide a deeper understanding of the actual physical environment, and play a key role in wireless network design, optimization and signal propagation simulation, especially in applications in complex indoor structural environments, such as smart buildings, security, indoor positioning and other fields.

[0163] In outdoor scenarios, scattered radio maps are usually composed of radio signal strength measurement data at different locations, which can reflect the signal coverage status in an area. Generating a global radio map through this scattered map has the following practical application significance:

[0164] The scattered radio map can be used to analyze the wireless signal coverage of a specified area (such as a city or a village). The global radio map can help network planners determine the signal strength distribution, interference source location, and signal dead spots, thereby optimizing the deployment location of base stations or wireless access points to ensure seamless network coverage and high-quality communication services;

[0165] The global radio map provides global data support for the propagation model of wireless signals, helping to identify coverage blind spots and weak signal areas, thereby improving overall network performance. Especially in large-scale cities or high-density areas, this map can help identify possible high-interference areas and optimize them;

[0166] When disasters occur (such as earthquakes, typhoons, etc.), global radio maps can help assess the extent of damage to communication facilities and the signal propagation conditions in the area, thereby providing guidance for emergency communications. Global radio maps updated in real time can ensure uninterrupted wireless communications in key areas.

[0167] In indoor environments, scattered radio maps usually involve wireless signal data collected from multiple locations, which can reflect the signal strength distribution at different locations in the space. Generating a global radio map in this way has the following application significance:

[0168] For large buildings (such as office buildings, shopping malls, hospitals, etc.), the global radio map obtained through the scattered radio map can help designers optimize the deployment location of wireless access points and improve the signal coverage and performance of the wireless network. Especially in high-rise buildings or environments with multiple partitions and walls, the global map can help predict signal attenuation and interference and optimize the layout;

[0169] Through the global radio map, combined with indoor positioning technology, the movement trajectory of people and objects can be accurately monitored. Especially for indoor positioning systems (such as smart offices and smart warehouses), this map provides higher precision and reliability, and improves the accuracy of the positioning system;

[0170] The distribution map of wireless signal strength can provide support for indoor smart device management. For example, by combining the global radio map and sensor data, smart buildings can automatically adjust temperature, lighting and other equipment to improve comfort and energy efficiency. At the same time, this map is also helpful for security monitoring, real-time monitoring of personnel distribution and movement to ensure safety.

[0171] In outdoor environments, object distribution maps usually represent the spatial layout of buildings, obstacles, vegetation, etc. in a geographical area. By combining object distribution maps with switching models, more accurate global radio maps can be generated. Its practical application significance includes:

[0172] The distribution map of outdoor objects (such as buildings, bridges, trees, etc.) directly affects the propagation of radio waves. The global radio map generated by the object distribution map can more accurately simulate the propagation path and attenuation of signals between different objects, thereby optimizing the location of base stations and network layout, and improving the coverage of communication networks;

[0173] The global radio map helps analyze the impact of the environment on wireless signals. For example, areas with dense buildings may cause signal obstruction, while areas with many trees may attenuate signal propagation. By combining the global map generated by the object distribution map, these factors can be evaluated and adjusted accordingly to ensure the stability of network services;

[0174] The global radio map generated by the object distribution map can identify sources of signal interference and propagation obstacles, helping wireless network planners reduce interference, avoid coverage blind spots, and improve communication quality.

[0175] In indoor scenarios, the object distribution map represents the layout of obstacles such as walls, furniture, and equipment in the building. By inputting this object information into the switching model, a global radio map can be generated. Its practical application significance includes:

[0176] Object distribution maps can help identify and predict indoor wireless signal attenuation, such as how walls, floors, furniture, etc. affect signal propagation. Through the global radio map, the layout of wireless access points is optimized to ensure comprehensive and stable wireless signal coverage in complex indoor environments (such as multi-story office buildings or commercial buildings).

[0177] For multi-story buildings, object distribution maps can help evaluate the performance of signal penetration between different floors, thereby helping network designers adjust the deployment of wireless devices on each floor. Through accurate global radio maps, the signal strength distribution of each floor can be determined to ensure seamless coverage across floors.

[0178] Combined with the object distribution map, the global radio map can provide more accurate indoor positioning services, especially in complex indoor environments (such as hospitals, airports, shopping malls, etc.). This precise positioning capability can realize automatic equipment control, such as the adjustment of smart devices such as air conditioners and lighting, while supporting accurate tracking of people and objects, improving the intelligence level of indoor environments.

[0179] There are many methods for obtaining outdoor object distribution maps and indoor object distribution maps. The specific method to be selected depends on factors such as the required accuracy, environmental characteristics, and available resources. Common methods include:

[0180] Outdoor building distribution map: Geographic Information System (GIS) data, remote sensing technology (satellite images, drones, lidar), building design drawings, etc.

[0181] Indoor object distribution map: building floor plan, laser scanning, indoor positioning technology, robots and drones, AR / VR technology, etc.

[0182] In outdoor environments, obtaining scattered radio maps usually requires actual radio signal measurements and data collection. Challenges in outdoor scenarios usually come from the impact of buildings, trees, and terrain undulations on signal propagation. Therefore, when obtaining scattered radio maps in outdoor environments, the following techniques and tools are usually required:

[0183] (1) Mobile measurement

[0184] Mobile measurements typically use vehicle-mounted systems, unmanned aerial vehicles (UAVs), or handheld devices with radio measurement equipment (such as Wi-Fi, LTE, 5G, or dedicated wireless bands) to measure signal strength.

[0185] Data collection: Install a radio receiver (such as a Wi-Fi module, LTE base station, or dedicated frequency band receiver) on a vehicle or drone, drive or fly along a predetermined route, and record data such as signal strength, signal-to-noise ratio (SNR), and multipath effect at each measurement point in real time.

[0186] Data Analysis: The collected data usually includes multiple signal parameters such as RSSI (Received Signal Strength Indicator), Signal-to-Noise Ratio (SNR), etc. These data are processed in Geographic Information System (GIS) and generate scattered radio maps.

[0187] Application:

[0188] On-vehicle measurement: Vehicles are equipped with radio signal receivers (such as Wi-Fi, LTE, 5G, etc.) and drive along the road, recording data such as signal strength at each location to generate a radio coverage map of the entire city or region.

[0189] Drone surveying: Drones carry surveying equipment and can cover large areas, especially in complex terrains such as cities or mountainous areas, providing detailed scattered radio data.

[0190] (2) Static base station measurement

[0191] Static base station measurements involve installing multiple radio receivers at fixed locations (e.g., Wi-Fi hotspots, LTE base stations, 5G base stations, etc.) and periodically measuring signal strength at these locations.

[0192] Data collection: By setting up multiple receiving sites, the radio signal strength at different locations is collected. The measuring equipment can include mobile phones, laptops, special radio frequency scanners, etc.

[0193] Data processing: By aggregating and analyzing the measurement data at different locations, a scattered radio map is formed to show the signal strength distribution at each location.

[0194] (3) Remote sensing measurement and satellite / UAV images

[0195] In some special cases, the accuracy of scattered radio maps can be further improved by combining remote sensing techniques (such as satellite or drone imagery) with signal measurement data.

[0196] Remote sensing and satellite imagery: High-resolution images acquired by remote sensing satellites or drones, combined with radio signal strength data collected on the ground, can better estimate the actual conditions of radio propagation.

[0197] The indoor environment is more complex than the outdoor environment, and signal propagation will be affected by obstacles such as walls, furniture, and electrical appliances. The method of obtaining scattered radio maps in indoor scenes is similar to that in outdoor scenes, but there are more factors to consider. The following are common methods:

[0198] (1) Indoor measurements based on radio equipment

[0199] Radio maps in indoor scenarios mainly rely on installing radio devices (such as Wi-Fi routers, Bluetooth beacons, UWB positioning systems, etc.) inside buildings and measuring signal strength at different locations.

[0200] Wi-Fi / Bluetooth measurements: Place Wi-Fi routers or Bluetooth beacons in different rooms, floors, or areas, and use a smartphone or dedicated measurement equipment to collect data such as signal strength (RSSI) and signal-to-noise ratio.

[0201] Data processing and visualization: Combine the radio parameters of each measurement location with its spatial coordinates and use visualization tools (such as Matlab, GIS software, etc.) to generate an indoor scattered radio map.

[0202] Application:

[0203] Wi-Fi signal measurement: In an indoor environment, use a Wi-Fi device to measure signal strength. You can draw a signal strength distribution map at different locations to help analyze wireless network coverage.

[0204] Bluetooth Beacons and UWB: Bluetooth or ultra-wideband (UWB) technology can also be used for precise indoor location and signal strength measurements, particularly for scattered radio map generation for positioning systems.

[0205] (2) Active and passive measurement

[0206] Measurements in indoor environments can be done using either active or passive methods.

[0207] Active measurement: By installing a radio transmitting device (such as a Wi-Fi router or Bluetooth device) indoors, the signal is actively generated and the signal strength at different locations is measured by a mobile device or receiver.

[0208] Passive measurement: By using existing radio signal sources (such as commercial Wi-Fi, Bluetooth devices, etc.), signal strength data is collected at different locations indoors. This method does not require additional radio transmission equipment and is applicable to existing network environments.

[0209] (3) RFID and UWB positioning systems

[0210] RFID (Radio Frequency Identification) and UWB positioning systems are technologies commonly used for precise indoor positioning. They can provide accurate signal strength data and are particularly suitable for large-scale or complex indoor environments.

[0211] RFID: Install RFID tags at every strategic location (such as shelves, doors, walls, etc.), use readers to collect signal strength data, and generate scattered radio maps.

[0212] UWB positioning system: UWB (ultra-wideband) technology can achieve centimeter-level positioning accuracy through very high frequencies and short pulse signals, which is particularly suitable for drawing scattered radio maps in complex indoor environments.

[0213] (4) Drones and robots

[0214] In complex indoor environments, using drones or ground robots to collect signals is an efficient way.

[0215] Drones: Drones can fly to specific locations indoors, carrying radio equipment to collect signal data, and are suitable for large-scale indoor environments (such as warehouses, factories, exhibition centers, etc.).

[0216] Robot: Indoor mobile robots can perform signal measurements in different areas, collect radio signal data in real time, and generate corresponding scattered radio maps based on the robot's location.

[0217] Example 4

[0218] The present invention provides a method for generating an object distribution map, which inputs a scattered radio map of a specified area into the switching model of the radio map and the object distribution map described in Example 2 to obtain the object distribution map of the specified area.

[0219] In outdoor scenarios, this method helps identify and analyze the layout of large-scale objects such as buildings, roads, and green spaces, and optimizes urban wireless network planning, communication signal propagation simulation, interference management, and disaster emergency response.

[0220] In indoor scenarios, by obtaining object distribution maps, it is possible to optimize indoor wireless network design, improve positioning system accuracy, and enhance automated management and security of smart buildings.

[0221] In outdoor environments, scattered radio maps are usually based on signal strength measurement data at multiple locations. These signal data can reflect the propagation of radio waves in an area, including signal strength, attenuation, and interference. Through these signal data, the distribution of objects in the area (such as buildings, roads, trees, etc.) can be inferred. Its practical application significance includes:

[0222] Signal attenuation in outdoor scattered radio maps can help identify and locate different objects (such as buildings, transportation facilities or large obstacles). The presence of these objects will affect the propagation of wireless signals, thereby changing the signal strength distribution. By inferring the scattered data, an accurate object distribution map can be generated, which helps to evaluate the role of buildings or other obstacles in radio wave propagation.

[0223] By obtaining object distribution maps, urban planners can understand the location and layout of large objects such as buildings, roads, and parks, and optimize the deployment of urban wireless networks. For example, reverse calculation of the distribution of buildings in a city can provide reference data for future base station construction and wireless network expansion.

[0224] The object distribution map provides detailed information about terrain features such as buildings, roads, and green spaces, making the radio wave propagation model more accurate. Through this map, the radio signal propagation simulation can be optimized, and the signal changes under different object distributions can be predicted, thereby optimizing the base station location, frequency planning, and network design, and improving communication quality.

[0225] In the event of a disaster (such as a fire or earthquake), the object distribution map can help assess which areas may be blocked and help emergency personnel determine the transmission of signals. Based on the scattered radio data, the object distribution map obtained in reverse can provide guidance to the rescue team and help quickly develop the best action plan.

[0226] In indoor environments, scattered radio maps are usually based on wireless signal strength measurements at different locations, reflecting the radio wave propagation within a room, floor, or building. By analyzing this data, the distribution of indoor objects such as walls, furniture, and equipment can be inferred. Its practical application significance includes:

[0227] Scattered radio maps can reflect different areas of indoor space, such as the layout of walls, partitions, and large furniture, through changes in wireless signals. Through these signal data, an accurate object distribution map can be generated to help understand the arrangement of objects in each room or area. This is of great significance for indoor wireless network optimization and layout design of indoor positioning systems.

[0228] In indoor environments, objects (such as walls, pillars, furniture, etc.) can affect signal propagation, causing signal attenuation and interference. Through the scattered radio map, the object distribution map obtained by reverse calculation can help predict and analyze the signal attenuation indoors, optimize the layout of wireless access points, and improve signal coverage and network stability.

[0229] In large commercial venues, hospitals, airports and other environments, object distribution maps can be combined with indoor positioning systems to provide accurate positioning data. The reversely calculated object distribution map can help better understand the path of signal propagation, thereby achieving more accurate personnel positioning and item tracking.

[0230] In smart buildings, object distribution maps can support the adjustment of automated equipment, such as the control of smart lighting, air conditioning, curtains, etc. The combination of accurate object distribution maps and scattered radio maps can provide smart buildings with more efficient and accurate management methods, improving the comfort and energy efficiency of living or office environments.

[0231] In the field of indoor security, object distribution maps can help optimize the layout of monitoring equipment and ensure that monitoring covers all important areas. Object distribution maps obtained through scattered radio maps can help detect and predict possible hidden areas and provide important basis for the optimization of security systems.

[0232] Example 5

[0233] The present invention provides a computer storage medium, a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method described in the above technical solution is implemented.

[0234] Example 6

[0235] The present invention also provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method described in the above technical solution by executing the computer instructions.

[0236] In order to verify the effectiveness of the proposed method, this specific embodiment is trained and tested on the simulation dataset RadioMapSeer and the measured dataset RSRPSet Urban for outdoor scenes, namely urban areas.

[0237] RadioMapSeer provides 700 building distribution maps from 6 different cities downloaded from OpenStreetMap. Each city map is set up with 80 base stations with omnidirectional antennas. Through Intelligent Ray Tracing (IRT) technology, a high-precision radio map corresponding to the location of different base stations in each city is obtained.

[0238] The RadioMapSeer dataset consists of 700 maps, each with 80 transmitter locations, and the corresponding simulated radio maps. The simulated data is generated using an intelligent ray tracing method based on 2D interactions of rays with geometry. City maps are taken from OpenStreetMap for cities such as Ankara, Berlin, Glasgow, London, and Tel Aviv.

[0239] Specifically, the simulation dataset forms a set of radio maps and a set of building distribution maps through the following steps:

[0240] 1. Data Source

[0241] Building distribution map: City maps downloaded from OpenStreetMap contain the location information and distribution of buildings. Each city map provides geometric information of buildings.

[0242] 2. Base station settings

[0243] Base station configuration: 80 base stations with omnidirectional antennas are set up on each city map. These base stations are located at different locations to cover the city area and provide basic signals for radio map generation.

[0244] 3. Intelligent ray tracing technology

[0245] High-precision radio map generation: Using intelligent ray tracing (IRT) technology, the signal propagation path in the urban environment is calculated by ray propagation simulation of each base station location. This includes taking into account factors such as signal reflection, diffraction and attenuation, thereby generating the signal strength of each base station at different locations.

[0246] 4. RSRP value calculation

[0247] Radio signal strength: The simulated ray tracing results are used to obtain the reference signal received power (RSRP) value at each location. These RSRP values ​​reflect the signal coverage of each base station at different locations in the city.

[0248] 5. Convert to pixel values

[0249] Radio map pixelation: The obtained RSRP values ​​are converted into pixel values ​​at the corresponding locations, ranging from 0 to 1. These pixel values ​​can represent the strength of the radio signal on the map in grayscale or color.

[0250] 6. Data cleaning and standardization

[0251] Normalization: Each radio map is processed using the maximum and minimum normalization method to ensure that the pixel values ​​are on the same scale to facilitate subsequent analysis and comparison.

[0252] 7. Final Output

[0253] Radio map and building distribution map: The resulting radio map shows the spatial distribution of the signal strength of each base station, while the building distribution map reflects the geometric distribution of buildings in the city. The combination of the two can analyze the impact of buildings on radio signal propagation.

[0254] The measured dataset RSRPSet Urban contains the RSRP of 415,244 signal receiving points measured in 181 densely populated urban communication cells. The electronic environment map of the cell in the dataset is obtained by converting the coordinates, altitude and building height in the RSRP dataset. According to the location of the cell base station, a 320m × 320m area along the antenna direction is gridded to represent the effective coverage of the cell base station. Each base station in the dataset is fixed at the grid (0, 32). The size of the cell electronic environment map is 64×64, and each pixel represents a 5m × 5m grid.

[0255] Specifically, the measured data set is formed into a radio map set and a building distribution map set through the following steps:

[0256] 1. Data Collection

[0257] Signal reception point data: RSRP values ​​of 415,244 signal reception points are extracted from the measured data set RSRPSet Urban. These data points are distributed in 181 urban communication cells and reflect the radio signal strength at different locations.

[0258] 2. Coordinates and environmental parameters

[0259] Coordinates and environmental characteristics: The data for each signal receiving point includes its coordinates, altitude, and corresponding building height. This information is used to build an electronic environmental map to help understand the actual environment in which the signal propagates.

[0260] 3. Base station location and coverage

[0261] Base station positioning: According to the location of the base station in the cell, a 320m x 320m coverage area is set. This area represents the effective coverage of the base station and provides the spatial basis for the generation of the radio map.

[0262] 4. Grid processing

[0263] Regional gridding: The coverage area of ​​320m x 320m is gridded to form a 64 x 64 electronic environment map, with each pixel representing an actual ground area of ​​5m x 5m. This subdivision makes the distribution of signal strength more accurate.

[0264] 5. Signal Strength Mapping

[0265] Radio signal mapping: The collected RSRP values ​​are mapped to the corresponding grid cells. Based on the RSRP value of the receiving point, the signal strength of each pixel is calculated, so that a radio map reflecting the signal strength distribution can be generated.

[0266] 6. Building distribution map generation

[0267] Building height and location: Based on the existing coordinates and building height information, the location information of the building can be marked on the electronic environment map to form a building distribution map. This graph can show the spatial distribution of buildings and their impact on signal propagation.

[0268] 7. Final Output

[0269] Radio map and building distribution map: The resulting radio map shows the signal strength of each grid cell, while the building distribution map shows the specific location and height of buildings in the city. The combination of the two makes it easier to analyze the shielding and reflection effects of buildings on radio signal propagation.

[0270] The simulation dataset RadioMapSeer and the measured dataset RSRPSet Urban are used to generate corresponding non-aligned training sets and corresponding validation sets for training the switching model of radio maps and object distribution maps. The validation set is used to generate a scattered radio map, which is input into the trained switching model to output a global radio map and an object distribution map to evaluate the radio map prediction performance and environment map reconstruction performance of the switching model.

[0271] This embodiment uses the existing technology (Radiocycle) and the ablation test (RadioYnet w / o RM) for comparison.

[0272] The existing technology RadioCycle receives power data of some reference signals collected in the cell, and proposes a wireless propagation model - RadioCycle by combining CycleGAN and U-Net. The mutual conversion between the cell's environmental map and radio map is regarded as a dual task, and the environmental map and radio map are predicted at the same time. The relevant papers are as follows:

[0273] Y.Zheng, T.Zhang, C.Liao, J.Wang, S.Liu, "RadioCycle: Deep Dual Learningbased Radio Map Estimation," KSII Transactions on Internet and InformationSystems, vol.16, no.11, pp.3780-3797, 2022.DOI:10.3837 / tiis.2022.11.017.

[0274] The ablation experiment (RadioYnet w / o RM) adopts the RadioYnet model proposed in the present invention, but only trains the first generator F.

[0275] Figure 4-Figure 7 In the figure, the horizontal axis represents the percentage of RSRP coverage area collected in the area represented by the scattered radio map used as input in the validation set. The percentage specifically refers to the percentage of the number of grids where RSRP can be collected compared to the total number of grids in the area after the area is gridded.

[0276] Figure 4 and Figure 5 In the figure, the vertical axis represents the root mean square error (RMSE). Figure 6 and Figure 7 In the figure, the vertical axis represents the intersection over union ratio, i.e. IOU (Intersection Over Union).

[0277] like Figure 4 As shown, in terms of radio map prediction performance of the simulation data set, the root mean square error of the present invention (RadioYnet) is significantly smaller than that of the prior art (Radiocycle).

[0278] like Figure 5 As shown, in terms of the radio map prediction performance of the measured data set, the root mean square error of the present invention (RadioYnet) is significantly smaller than that of the prior art (Radiocycle).

[0279] like Figure 6As shown, in terms of the environmental map reconstruction performance of the simulation data set, the IOU of the present invention (RadioYnet) is significantly greater than that of the prior art (Radiocycle) and the ablation test (RadioYnet w / o RM), indicating that the cyclic training method of the present invention is stronger than single training in the task of environmental reconstruction.

[0280] like Figure 7 As shown in the figure, in terms of the environmental map reconstruction performance of the measured data set, the IOU of the present invention (RadioYnet) is significantly greater than that of the prior art (Radiocycle) and the ablation test (RadioYnet w / o RM), indicating that the cyclic training method of the present invention is stronger than single training in the task of environmental reconstruction.

[0281] In summary, the performance of the present invention is better than the existing prior art.

[0282] like Figure 8 As shown, from left to right, a verification sample scatter radio map in the simulation dataset, the actual object distribution map, the object distribution map output by the model, the global radio map output by the model, and the actual global radio map are shown.

[0283] like Fig. 9 As shown in the figure, from left to right, a scatter radio map of a verification sample in the actual dataset, the actual object distribution map, the object distribution map output by the model, the global radio map output by the model, and the actual global radio map are shown.

[0284] Obviously, the present invention can effectively restore the global radio map and object distribution map through the scattered radio map. The present invention can estimate the RSRP distribution of the remaining (100-N)% area where no signal transceiver equipment is deployed by randomly deploying signal transceiver equipment in N percent of the area in the specified area; or estimate the object distribution of the entire area by randomly deploying signal transceiver equipment in N percent of the area in the specified area.

[0285] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0286] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0287] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0288] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0289] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

[0290] The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

Claims

1. A training method for a switching model of a radio map and an object distribution map, characterized in that: The following steps are involved: Generate non-aligned training sets: construct training data containing a set of actual object distribution maps and a set of actual global radio maps; Randomly generate signal transceiver device distribution information: generate random signal transceiver device distribution information for each training round of the first generator and the second generator; The training process is repeated until the model reaches the desired accuracy: Based on any actual global radio map in the actual global radio map set, generate a scattered radio map by sampling; Inputting the scattered radio map and the signal transceiver device distribution information into a first generator to obtain an estimated object distribution map; Inputting the estimated object distribution map and the signal transceiver device distribution information into a second generator to obtain an estimated global radio map; Inputting any actual object distribution map in the actual object distribution map set and the signal transceiver device distribution information into a second generator to obtain a simulated global radio map; Inputting the simulated global radio map and the signal transceiver device distribution information into the first generator to obtain a simulated object distribution map; Calculate the loss function and update the model parameters: Calculate the consistency loss function between the estimated global radio map and the actual global radio map; Calculate the generative adversarial loss function between the simulated global radio map and the actual global radio map; Calculate the consistency loss function between the simulated object distribution map and the actual object distribution map; Calculate the generative adversarial loss function of the estimated object distribution map and the actual object distribution map; based on the above loss function, update the parameters of the first generator and the second generator.

2. The method for training a switching model of a radio map and an object distribution map according to claim 1, characterized in that: During a single round of training of the first generator and the second generator, the distribution information of the signal transceiver devices used is the same.

3. The method for training a switching model of a radio map and an object distribution map according to claim 1, characterized in that: The area types corresponding to the actual object distribution map set and the actual global radio map set are the same as the area types of the designated areas; the area types include urban outdoor areas and indoor areas.

4. The method for training a switching model of a radio map and an object distribution map according to claim 1, characterized in that: The consistency loss function between the estimated global radio map and the actual global radio map uses the mean square error loss function.

5. The method for training a switching model of a radio map and an object distribution map according to claim 1, characterized in that: The consistency loss function between the simulated object distribution map and the actual object distribution map adopts the dice loss function.

6. A switching model between a radio map and an object distribution map, characterized in that: The model is obtained by training through the training method according to any one of claims 1 to 5, comprising: A first generator: used for generating an object distribution map based on a scattered radio map; The second generator is used to generate a global radio map based on the object distribution map and the signal transceiver device distribution information.

7. A method for generating a radio map, characterized in that: The object distribution map and signal transceiver equipment distribution information of the designated area, or the scattered radio map is input into the switching model of the radio map and the object distribution map as claimed in claim 6 to obtain a global radio map of the designated area.

8. A method for generating an object distribution map, characterized in that: The scattered radio map of the designated area is input into the switching model of the radio map and the object distribution map as claimed in claim 6 to obtain the object distribution map of the designated area.

9. A computer storage medium, a computer readable storage medium, having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to claim 7 or 8 is implemented.

10. An electronic device, characterized in that: The electronic device includes: a processor for executing a computer program stored on the device; A memory for storing the computer program; Wherein, when the computer program is executed by a processor, the method described in claim 7 or 8 is implemented.

Images