A three-dimensional radio map construction method and related device

Abstract

This invention provides a method and related equipment for constructing a three-dimensional radio map. The method processes radio maps under different scenarios, operating frequency bands, and receiving altitudes to obtain training radio map data and basic environmental information data. It then pre-trains a general-purpose radio map model using the training radio map data to obtain an initial model, and uses the basic environmental information data to provide cueed learning to obtain a radio map construction model. Finally, it inputs sparse measurement data of the target area, a three-dimensional building map, the target frequency band, and the target altitude into the radio map construction model to construct a map, thereby improving the robustness of the three-dimensional radio map construction.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of wireless communication and artificial intelligence, and in particular to a method and related equipment for constructing three-dimensional radio maps. Background Technology

[0002] As a visual representation of the spatial distribution of wireless signals, radio maps can accurately reflect the geographical distribution patterns of key indicators such as radio frequency signal functions, signal-to-noise ratio, and channel characteristics. They are a core technological support for optimizing spectrum utilization efficiency and improving network performance and user experience.

[0003] With the evolution of wireless communication technology, traditional two-dimensional single-band radio maps can no longer meet the needs: on the one hand, the development of integrated air-space-ground networks requires radio maps to have three-dimensional spatial representation capabilities to adapt to emerging applications such as drone communication and satellite base station deployment; on the other hand, the popularization of multi-band heterogeneous networks requires maps to be able to simultaneously depict the propagation characteristics of signals of different frequencies to support intelligent spectrum management.

[0004] Existing radio map construction methods have significant limitations: 1. Insufficient generalization ability: Traditional physical models (such as log distance path loss models) and interpolation methods (such as radial basis functions and inverse distance weighting) rely on fixed assumptions or spatial correlations of sparse measurements, making it difficult to adapt to diverse scenarios such as urban areas, rural areas, indoor areas, and mountainous areas; existing deep learning-based solutions (such as Radio U-Net and RME GAN) are mostly trained for specific scenarios (such as urban areas) and lack a general adaptation mechanism across scenarios and frequency bands.

[0005] 2. Lack of dimensional coverage: Most existing models focus on a single frequency and a fixed altitude, without fully considering the differences in propagation loss between different frequency bands (such as strong diffraction at low frequencies and line-of-sight propagation characteristics at high frequencies), i.e., the impact of vertical dimension (height) on signal distribution, and thus cannot support the construction of multi-band three-dimensional radio maps.

[0006] 3. Inefficient use of auxiliary information: Auxiliary information such as environmental description, frequency, and altitude has a key impact on signal propagation, but existing solutions have difficulty effectively integrating this abstract information into the model training and inference process, resulting in a significant drop in model performance in unseen scenarios (untrained scenarios, frequency bands, or altitudes). Summary of the Invention

[0007] This invention provides a method and related equipment for constructing three-dimensional radio maps, the purpose of which is to improve the robustness of three-dimensional radio map construction.

[0008] To achieve the above objectives, the present invention provides a method for constructing a three-dimensional radio map, comprising: Step 1: Acquire training radio map data and basic environmental information data under different scenarios, different operating frequency bands, and different receiving altitudes; Step 2: Pre-train the general radio map model using the training radio map data to obtain the initial model, and use basic environmental information data to guide the initial model to learn, thus obtaining the radio map construction model. Step 3: Input the sparse measurement data of the target area, the 3D building map, the target frequency band and the target height into the radio map construction model to construct the map and obtain the 3D radio map of the target area.

[0009] Furthermore, the general radio map big model includes an encoder for extracting shallow environmental information features from the training radio map data, a cue network for converting basic environmental information data into cue embeddings, a deep U-Net backbone network for fusing shallow environmental information features with cue embeddings, and a decoder for converting high-level features output by the deep U-Net backbone network into radio maps. The encoder's input and the cueing network are both inputs to the general radio map model, used to access training radio map data and basic environmental information data respectively during the training phase. The output of the encoder and the output of the cue network are both connected to the input of the deep U-Net backbone network, and the output of the deep U-Net backbone network is connected to the input of the decoder. The decoder's output is the output of the general radio map large model, used to output the radio map.

[0010] Furthermore, the cue network includes a building density cue subnetwork, a frequency cue subnetwork, and a height cue subnetwork; The building density cueing subnetwork is used to process the 3D building environment information in the basic environmental information data to obtain the building cueing embedding; The frequency cueing subnetwork is used to process the frequency propagation knowledge in the basic environmental information data to obtain the frequency cueing embedding; The height cue subnetwork is used to process the height information in the basic environmental information data to obtain the height cue embedding.

[0011] Furthermore, the building density cue subnetwork, frequency cue subnetwork, and height cue subnetwork all include a feature extraction module connected in sequence and a memory pool for generating cue embeddings.

[0012] Furthermore, the deep U-Net backbone network has a symmetrical structure, including: An encoding module used to extract hierarchical features through downsampling; A decoding module used to reconstruct high-resolution representations through upsampling; A context modeling module consisting of a first residual block, a scaled dot product self-attention unit, and a second residual block connected in sequence; The output of the encoding module is connected to the input of the decoding module through the context modeling module.

[0013] Furthermore, the scaled dot product self-attention unit includes a first linear layer, a second linear layer, a third linear layer, a scaled dot product attention layer, a connection layer, and a fourth linear layer; The query vector, key vector, and value vector are obtained by utilizing the features of the input. Input the query vector, key vector, and value vector into the first linear layer, the second linear layer, and the third linear layer, respectively. The outputs of the first, second, and third linear layers are all connected to the input of the scaled dot product attention layer. The output of the scaled dot product attention layer is connected to the output of the connection layer. The input of the connection point is connected to the output of the fourth linear layer. The output of the linear layer is the output of the scaled dot product self-attention unit.

[0014] The present invention also provides a three-dimensional radio map construction apparatus, comprising: The acquisition module is used to acquire training radio map data and basic environmental information data under different scenarios, different operating frequency bands, and different receiving altitudes. The training module is used to pre-train a general radio map model using training radio map data to obtain an initial model, and to guide the initial model to learn using basic environmental information data to obtain a radio map construction model. The construction module is used to input sparse measurement data, 3D building maps, target frequency bands, and target heights of the target area into the radio map construction model to construct a map and obtain a 3D radio map of the target area.

[0015] The present invention also provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a three-dimensional radio map construction method.

[0016] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a method for constructing a three-dimensional radio map.

[0017] The above-described solution of the present invention has the following beneficial effects: This invention acquires training radio map data and basic environmental information data under different scenarios, operating frequency bands, and receiving altitudes. It pre-trains a general-purpose radio map model using the training radio map data to obtain an initial model, and then guides the initial model's learning using the basic environmental information data to obtain a radio map construction model. Sparse measurement data of the target area, 3D building maps, the target frequency band, and the target altitude are input into the radio map construction model to construct a map, resulting in a 3D radio map of the target area. Compared with existing technologies, this invention pre-trains a general-purpose radio map model using radio maps under different scenarios, operating frequency bands, and receiving altitudes, enabling the model to learn the essential laws of signal propagation, thereby improving the robustness of 3D radio map construction. Simultaneously, guiding the initial model's learning using basic environmental information data allows the model to quickly adapt to new data distributions without retraining.

[0018] Other beneficial effects of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating an embodiment of the present invention; Figure 2 This is a schematic diagram of the framework of an embodiment of the present invention; Figure 3 This is a schematic diagram of the prompting network structure in an embodiment of the present invention; Figure 4 The graph shows the performance analysis results of the embodiments of the present invention at different sampling rates; Figure 5 The figure shows the performance analysis results of the embodiment of the present invention under noise conditions; Figure 6 This is a schematic diagram of the structure of the three-dimensional radio map construction device in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of the terminal device in an embodiment of the present invention. Detailed Implementation

[0020] To make the technical problems, solutions, and advantages of this invention clearer, a detailed description will be provided below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0021] In the description of this invention, it should be noted that the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0022] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a locking connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0023] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0024] This invention addresses existing problems by providing a method and related equipment for constructing three-dimensional radio maps.

[0025] like Figure 1 As shown, an embodiment of the present invention provides a method for constructing a three-dimensional radio map, including: Step 1: Acquire training radio map data and basic environmental information data under different scenarios, different operating frequency bands, and different receiving altitudes; Step 2: Pre-train the general radio map model using the training radio map data to obtain the initial model, and use basic environmental information data to guide the initial model to learn, thus obtaining the radio map construction model. Step 3: Input the sparse measurement data of the target area, the 3D building map, the target frequency band and the target height into the radio map construction model to construct the map and obtain the 3D radio map of the target area.

[0026] Specifically, step 1 includes: The initial training radio map data is obtained or generated under different scenarios, operating frequency bands, and receiving altitudes. Different scenarios include urban, rural, mountainous, lake, and grassland environments; different operating frequency bands are from 150MHz to 220GHz; and different receiving altitudes are 1.5m, 30m, and 200m. The radio map includes a 3D building layout, sparse measurement points and their locations, and signal strength values. The spatial resolution is 10m, and the size of a single map is 128×128, corresponding to an actual area size of 1.28km×1.28km. To facilitate model training, the received power values ​​of the initial training radio map neural network are normalized to [0,1] to obtain training radio map data and basic environmental information data, which include environmental description, frequency, and altitude.

[0027] Specifically, such as Figure 2As shown, the general radio map large model includes an encoder for extracting shallow features of environmental information from training radio map data, a cue network for converting basic environmental information data into cue embeddings, a deep U-Net backbone network for fusing shallow features of environmental information with cue embeddings, and a decoder for converting high-level features output by the deep U-Net backbone network into radio maps. The encoder's input and the cueing network are both inputs to the general radio map model, used to access training radio map data and basic environmental information data respectively during the training phase. The output of the encoder and the output of the cue network are both connected to the input of the deep U-Net backbone network, and the output of the deep U-Net backbone network is connected to the input of the decoder. The decoder's output is the output of the general radio map large model, used to output the radio map.

[0028] In this embodiment of the invention, the encoder, the deep U-Net backbone network, and the decoder form the backbone network of a general radio map model to learn general latent representations and capture common features of signal propagation under different scenarios, frequency bands, and altitudes, laying the foundation for cross-domain generalization. The prompt network language transforms basic environmental information data into prompt embeddings to dynamically adapt to different data distributions and achieve rapid adaptation to unknown scenarios, frequency bands, and altitudes.

[0029] In this embodiment of the invention, during the training process, sparse measurement points and 3D building layouts from the training radio map data are used as inputs to a general-purpose large-scale radio map model. The complete radio map from the training radio map data is simultaneously input as a supervision signal into the general-purpose large-scale radio map model. The AdamW optimizer and cosine annealing learning rate scheduling are used, with an initial learning rate of 5 × 10⁻⁶. -5 The cosine decays to 1×10 -6 During the pre-training phase, the backbone network parameters are optimized by minimizing the difference between the predicted and true values ​​using the root mean square error loss function. The loss function is as follows: ; in, For the predicted radio map, This is a real radio map.

[0030] Specifically, the encoder consists of standard convolutional layers used to transform sparse measurement points and 3D building layouts into shallow features of environmental information.

[0031] In this embodiment of the invention, the cue network is the key to achieving the model’s universality and adaptability. Its core is to transform basic environmental information data into cue embeddings that the model can understand and inject them into the pre-trained initial model to guide it in constructing radio maps under specific conditions.

[0032] Specifically, such as Figure 3 As shown, the cue network includes a building density cue subnetwork, a frequency cue subnetwork, and a height cue subnetwork; The building density cueing subnetwork is used to process the 3D building environment information in the basic environmental information data to obtain the building cueing embedding; The frequency cueing subnetwork is used to process the frequency propagation knowledge in the basic environmental information data to obtain the frequency cueing embedding; The height cue subnetwork is used to process the height information in the basic environmental information data to obtain the height cue embedding.

[0033] Specifically, the building density cue subnetwork, frequency cue subnetwork, and height cue subnetwork all include a feature extraction module connected in sequence and a memory pool for generating cue embeddings.

[0034] In this embodiment of the invention, three-dimensional building projection images at different heights are displayed. The input building density cue subnetwork captures environmental structures of varying coarseness and fineness through a feature extraction module composed of multi-scale convolutional layers, yielding building density features, expressed as: ; in, Indicates building density characteristics, This indicates multi-scale convolution processing; Building density features are input into a memory pool of building densities. Valuable knowledge is extracted from the memory pool by using the building density features as queries, resulting in a building cue embedding, expressed as: ; in, This indicates that the building prompt is embedded. This indicates the number of memory units in the memory pool. Indicates the first The contribution weight of each memory unit to the current input scenario. Represents the first in the memory pool key-value pairs This represents the softmax function, used for adaptive weight allocation; First, input the scalar frequency value into the frequency prompt subnetwork; Then, the Friis free space path loss model is used as a feature extraction module to transform the frequency values ​​into physically meaningful spatial path loss features. Then, based on the sparse measurement points, the N points with the strongest power are selected as "virtual transmitters," and their relative distance matrices to each point in space are calculated. And path loss, the calculation of path loss is expressed as follows: ; in, It is a unit conversion constant; By combining the contributions of all virtual transmitters, a frequency-dependent spatial signal intensity feature matrix is ​​generated, i.e. ; in, It is the first The received signal strength of a virtual transmitter; Finally, the spatial signal intensity feature matrix is ​​encoded using a convolutional layer and then input into the memory pool. Valuable memory knowledge is extracted from the memory pool by using the encoded spatial signal intensity feature matrix as a query, resulting in the frequency cue embedding, expressed as: ; scalar height value And the building information input height prompt subnetwork for that height level; Then, the additional path loss caused by altitude changes is calculated based on the formula for generating the spatial signal strength feature matrix, and this loss is integrated into the calculation of frequency characteristics. ; Next, multi-scale convolutional layers are used as feature extraction modules to extract height-related features at specific height layers, as expressed in the following expression: ; Finally, highly relevant features are input into a memory pool. Valuable memory knowledge is extracted from the memory pool by using these highly relevant features as queries, resulting in a highly cueing embedding, expressed as: ; in, This indicates a high level of embedding.

[0035] In an embodiment of the invention, a building density memory pool network It can be represented as: ; in, It is the number of memory units. These are key-value pairs in the memory pool, all of them. and These are learnable parameters.

[0036] Specifically, the deep U-Net backbone network has a symmetrical structure, including: An encoding module used to extract hierarchical features through downsampling; A decoding module used to reconstruct high-resolution representations through upsampling; A context modeling module consisting of a first residual block, a scaled dot product self-attention unit, and a second residual block connected in sequence; The output of the encoding module is connected to the input of the decoding module through the context modeling module.

[0037] In this embodiment of the invention, the encoding module consists of two residual blocks and a downsampling layer, wherein the downsampling layer is disposed between the two residual blocks. The decoding module consists of two residual blocks and an upsampling layer, wherein the upsampling layer is disposed between the two residual blocks. The output of the second residual block in the encoding module is connected to the input of the first residual block in the decoding module through a context modeling module. All other structures in the encoding module except the second residual block are connected to the decoding module in a skip connection.

[0038] In this embodiment of the invention, the deep U-Net backbone network is used to fuse shallow features of environmental information with building cue embedding, frequency cue embedding, and altitude cue embedding, thereby dynamically guiding the model to generate accurate radio maps adapted to specific environments, frequency bands, and altitudes.

[0039] Specifically, both the first residual block and the second residual module include multiple stacked convolutional layers, normalization layers, and activation function layers.

[0040] Specifically, the scaled dot product self-attention unit includes a first linear layer, a second linear layer, a third linear layer, a scaled dot product attention layer, a connection layer, and a fourth linear layer; The query vector, key vector, and value vector are obtained by utilizing the features of the input. Input the query vector, key vector, and value vector into the first linear layer, the second linear layer, and the third linear layer, respectively. The outputs of the first, second, and third linear layers are all connected to the input of the scaled dot product attention layer. The output of the scaled dot product attention layer is connected to the output of the connection layer. The input of the connection point is connected to the output of the fourth linear layer. The output of the linear layer is the output of the scaled dot product self-attention unit.

[0041] In this embodiment of the invention, the scaled dot product self-attention layer is used to calculate the dependencies of all spatial locations in the feature map, enhancing the model's ability to model the propagation patterns of long-distance signals. The formula is as follows: ; in, , , , These are the features of the input. , , These are learnable projection weights.

[0042] This invention embodiment conducts experiments based on the SpectrumNet dataset, which contains 3D radio map data for 11 different scene types, 5 frequency bands, and 3 elevations. Each region is 1.28km × 1.28km in size, with a spatial resolution of 10m (i.e., 128 × 128 grids). For input settings, approximately 1% of observation points are randomly and uniformly sampled from the complete radio map as sparse input, combined with the corresponding 3D building distribution information as prior environmental input. The training strategy uses a training set / validation set / test set ratio of 7:1:2. Under the above unified experimental settings, the method provided in this invention embodiment is compared with various baseline methods (including interpolation methods and deep learning methods), and MSE, NMSE, RMSE, and PSNR are used as evaluation metrics. The radio map construction performance comparison results are shown in Table 1 below: Table 1. Performance Comparison Results of Radio Map Construction

[0043] In Table 1 above, indicators marked with "↓" indicate that smaller values ​​result in better performance, while indicators marked with "↑" indicate that larger values ​​result in better performance. The bolded portion represents the best results, and the underlined portion represents the second-best results. Table 1 shows that on a large-scale dataset (SpectrumNet) containing 11 scenes, 5 frequency bands, and 3 heights, the performance of this embodiment significantly outperforms existing best methods in key metrics such as MSE, RMSE, NMSE, and PSNR. Based on the cross-regional scale generalization experiment setup, the performance results for different region sizes are shown in Table 2: Table 2 Performance test results for regions of different sizes

[0044] Specifically, in this embodiment of the invention, the model is first trained on data in a standard-scale region (1.28km × 1.28km), and then tested on regions of different spatial scales (e.g., 1km × 1km) to evaluate the model's generalization ability under varying spatial scale conditions. Based on inputs with different sparse sampling rates, the following results are obtained: Figure 4 Performance variation curves at different sampling rates.

[0045] Specifically, in this embodiment of the invention, under the same dataset and model structure, different input sampling rates are constructed to create input data with varying sparsity, including but not limited to: 0.5%, 1%, 2%, 5%, etc., based on different noise interference conditions, to obtain... Figure 5 The performance evaluation results under the noise conditions are shown.

[0046] Specifically, in this embodiment of the invention, noise interference of different intensities is introduced on the basis of the original sparse observation data to simulate the uncertainty in actual wireless measurement. The noise is modeled as Gaussian noise, and its intensity can be controlled by the noise variance. Under different noise conditions, the input sparse observation values ​​are perturbed.

[0047] From the above table 2, Figure 4 , Figure 5 It can be seen that, under harsh conditions such as low sampling rate (1%), measurement noise, and different region sizes, the embodiments of the present invention all exhibit more stable and superior performance than the baseline model.

[0048] This invention acquires training radio map data and basic environmental information data under different scenarios, operating frequency bands, and receiving altitudes. It pre-trains a general-purpose radio map model using the training radio map data to obtain an initial model, and then guides the initial model's learning using the basic environmental information data to obtain a radio map construction model. Sparse measurement data of the target area, 3D building maps, the target frequency band, and the target altitude are input into the radio map construction model to construct a map, resulting in a 3D radio map of the target area. Compared with existing technologies, this invention pre-trains a general-purpose radio map model using radio maps under different scenarios, operating frequency bands, and receiving altitudes, enabling the model to learn the essential laws of signal propagation and improving the robustness of 3D radio map construction. Simultaneously, guiding the initial model's learning using basic environmental information data allows the model to quickly adapt to new data distributions without retraining.

[0049] Corresponding to the three-dimensional radio map construction method described in the above embodiments, such as Figure 6 As shown, this embodiment of the invention also provides a three-dimensional radio map building device 100, which includes: The acquisition module 101 is used to acquire training radio map data and basic environmental information data under different scenarios, different operating frequency bands, and different receiving altitudes. Training module 102 is used to pre-train a general radio map model using training radio map data to obtain an initial model, and to guide the initial model to learn using basic environmental information data to obtain a radio map construction model. The construction module 103 is used to input sparse measurement data of the target area, three-dimensional building map, target frequency band and target height into the radio map construction model to construct a map and obtain a three-dimensional radio map of the target area.

[0050] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0051] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0052] This invention also provides a terminal device, such as... Figure 7 As shown, the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 7 The diagram shows only one processor, a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100. When the processor D100 executes the computer program D102, it implements the above-described three-dimensional radio map construction method.

[0053] The terminal device D10 can be a desktop computer, laptop, handheld computer, server, server cluster, or cloud server, etc. This terminal device may include, but is not limited to, a processor D100 and a memory D101. Those skilled in the art will understand that... Figure 7 This is merely an example of terminal device D10 and does not constitute a limitation on terminal device D10. It may include more or fewer components than shown in the figure, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0054] The processor D100 can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0055] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may be an external storage device of the terminal device D10, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal device D10. Furthermore, the memory D101 may include both internal and external storage units of the terminal device D10. The memory D101 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory D101 can also be used to temporarily store data that has been output or will be output.

[0056] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0057] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0058] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a three-dimensional radio map construction method.

[0059] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying the computer program code to a building device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0060] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for constructing a three-dimensional radio map, characterized in that, include: Step 1: Acquire training radio map data and basic environmental information data under different scenarios, different operating frequency bands, and different receiving altitudes; Step 2: Use the training radio map data to pre-train the general radio map large model to obtain an initial model, and use the basic environmental information data to guide the initial model to learn, thereby obtaining a radio map construction model; Step 3: Input the sparse measurement data of the target area, the 3D building map, the target frequency band and the target height into the radio map construction model to construct the map and obtain the 3D radio map of the target area.

2. The three-dimensional radio map construction method according to claim 1, characterized in that, The general radio map big model includes an encoder for extracting shallow environmental information features from the training radio map data, a cue network for converting basic environmental information data into cue embeddings, a deep U-Net backbone network for fusing shallow environmental information features with cue embeddings, and a decoder for converting high-level features output by the deep U-Net backbone network into a radio map. The input terminals of the encoder and the prompting network are both input terminals of the general radio map large model, used to access training radio map data and basic environmental information data respectively during the training phase; The output of the encoder and the output of the cueing network are both connected to the input of the deep U-Net backbone network, and the output of the deep U-Net backbone network is connected to the input of the decoder. The output of the decoder is the output of a general radio map large model, used to output radio maps.

3. The three-dimensional radio map construction method according to claim 2, characterized in that, The prompting network includes a building density prompting subnetwork, a frequency prompting subnetwork, and a height prompting subnetwork; The building density cueing subnetwork is used to process the three-dimensional building environment information in the basic environmental information data to obtain building cueing embeddings; The frequency cue subnetwork is used to process the frequency propagation knowledge in the basic environmental information data to obtain the frequency cue embedding; The height cue subnetwork is used to process the height information in the basic environmental information data to obtain the height cue embedding.

4. The three-dimensional radio map construction method according to claim 3, characterized in that, The building density cue subnetwork, the frequency cue subnetwork, and the height cue subnetwork each include a feature extraction module and a memory pool for generating cue embeddings, which are connected in sequence.

5. The three-dimensional radio map construction method according to claim 4, characterized in that, The deep U-Net backbone network has a symmetrical structure, including: An encoding module used to extract hierarchical features through downsampling; A decoding module used to reconstruct high-resolution representations through upsampling; A context modeling module consisting of a first residual block, a scaled dot product self-attention unit, and a second residual block connected in sequence; The output of the encoding module is connected to the input of the decoding module through the context modeling module.

6. The three-dimensional radio map construction method according to claim 5, characterized in that, The residual block consists of stacked convolutional layers, grouped normalization layers, and SiLU activation function layers.

7. The method for constructing a three-dimensional radio map according to claim 5, characterized in that, The scaled dot product self-attention unit includes a first linear layer, a second linear layer, a third linear layer, a scaled dot product attention layer, a connection layer, and a fourth linear layer; The query vector, key vector, and value vector are obtained by utilizing the features of the input. The query vector, the key vector, and the value vector are respectively input into the first linear layer, the second linear layer, and the third linear layer; The outputs of the first linear layer, the second linear layer, and the third linear layer are all connected to the input of the scaled dot product attention layer. The output of the scaled dot product attention layer is connected to the output of the connection layer. The input of the connection layer is connected to the output of the fourth linear layer. The output of the linear layer is the output of the scaled dot product self-attention unit.

8. A three-dimensional radio map construction device, characterized in that, include: The acquisition module is used to acquire training radio map data and basic environmental information data under different scenarios, different operating frequency bands, and different receiving altitudes. The training module is used to pre-train the general radio map model using the training radio map data to obtain an initial model, and to guide the initial model to learn using the basic environmental information data to obtain a radio map construction model. The construction module is used to input sparse measurement data of the target area, 3D building map, target frequency band and target height into the radio map construction model to construct a map and obtain a 3D radio map of the target area.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the three-dimensional radio map construction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the three-dimensional radio map construction method as described in any one of claims 1 to 7.

Images