Wireless signal loss prediction method, apparatus, device, medium, and program product

Abstract

The application provides a wireless signal loss prediction method and device, equipment, medium and program product, relates to the technical field of artificial intelligence, and is used for improving the accuracy of indoor space wireless signal loss prediction results. The specific technical scheme is as follows: obtaining a penetration loss cumulative value of a first path corresponding to each grid point in an indoor space; obtaining a reflection eigenvalue corresponding to each grid point; based on the penetration loss cumulative value of the first path corresponding to each grid point and the reflection eigenvalue corresponding to each grid point, predicting a loss value of the wireless signal in the indoor space. The application is applied to the scene of indoor space wireless channel modeling and prediction.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device, medium, and program product for predicting wireless signal loss. Background Technology

[0002] In the field of wireless communication, modeling and predicting the propagation of wireless signals is a widely concerned issue. In particular, the path loss experienced by wireless signals is the most basic feature required to analyze coverage area, interference power and other related indicators.

[0003] In related technologies, the propagation of wireless signals in an indoor environment can typically be simulated through physical simulation calculations, based on full-wave simulation and ray tracing techniques. Alternatively, statistical models can be used to describe the propagation loss distribution of wireless signals.

[0004] However, the former method is more limited, requiring a complete rerun of the simulated wireless signal propagation even for minor environmental changes, lacking flexibility. The latter, on the other hand, struggles to adapt to complex and changing indoor environments, resulting in lower accuracy in predicting wireless signal propagation loss. Summary of the Invention

[0005] This application provides a method, apparatus, device, medium, and program product for predicting wireless signal loss, which improves the accuracy of predicting the propagation loss of wireless signals.

[0006] In a first aspect, embodiments of this application provide a wireless signal loss prediction method, the method comprising: obtaining a cumulative penetration loss value of a first path corresponding to each grid point in an indoor space, wherein the first path corresponding to each grid point is a path from the grid point to a target signal transmitter; the cumulative penetration loss value of a first path is the sum of the penetration loss values ​​of the wireless signal emitted by the target signal transmitter through each obstacle on the first path; obtaining a reflection characteristic value corresponding to each grid point; the reflection characteristic value is determined based on the direct path loss value of the wireless signal to each grid point and the minimum reflection path loss value of the wireless signal to each grid point; and predicting the loss value of the wireless signal in the indoor space based on the cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point.

[0007] The technical solution provided in this application has at least the following beneficial effects: This solution fully considers the penetration loss value of wireless signals penetrating obstacles in indoor spaces and the reflection loss value of wireless signals reflecting in indoor spaces, so that wireless signal loss during the propagation process can be predicted by multiple loss values, thereby improving the accuracy of predicting wireless signal loss during the propagation process.

[0008] One possible implementation method, wherein obtaining the cumulative penetration loss value of the first path corresponding to each grid point in the indoor space includes: determining the penetration loss value corresponding to each obstacle among all obstacles traversed in the first path where the first grid point is located; the first grid point is one of the grid points; and performing cumulative calculation on the penetration loss value corresponding to each obstacle to determine the cumulative penetration loss value corresponding to the first path where the first grid point is located.

[0009] In another possible implementation, before obtaining the reflection feature value corresponding to each grid point, the method further includes: obtaining the first free space propagation loss value of the wireless signal to the first grid point; the first grid point is one of the grid points; adding the cumulative penetration loss value corresponding to the first path where the first grid point is located to the first free space propagation loss value to obtain the direct path loss value of the wireless signal to the first grid point.

[0010] In another possible implementation, before obtaining the reflection feature value corresponding to each grid point, the method further includes: obtaining the second free space propagation loss value of the wireless signal to the first reflection point; the first reflection point is one of the reflection points corresponding to the first grid point, and the first grid point is one of the grid points; the second free space propagation loss value, the cumulative penetration loss value corresponding to the second path where the first reflection point is located, and the reflection loss generated by the wireless signal to the first reflection point are added together to obtain the reflection path loss value of the first grid point; wherein, the second path where the first reflection point is located is the path between the first reflection point and the target signal transmitter, and the cumulative penetration loss value corresponding to a second path is the sum of the penetration loss values ​​of the wireless signal emitted by the target signal transmitter through each obstacle on the second path.

[0011] Another possible implementation, which predicts the loss value of the wireless signal in the indoor space based on the cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point, includes: inputting the cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point into the wireless signal loss prediction model; and outputting the predicted wireless signal loss value of each grid point in the indoor space.

[0012] Another possible implementation involves inputting the cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point into the wireless signal loss prediction model, including: inputting a first matrix and a second matrix into the wireless signal loss prediction model; wherein the first matrix is ​​constructed based on the position of each grid point in the indoor space and the cumulative penetration loss value of the first path corresponding to each grid point, and one element of the first matrix corresponds to the cumulative penetration loss value of the first path corresponding to one grid point; the second matrix is ​​constructed based on the position of each grid point in the indoor space and the reflection characteristic value corresponding to each grid point, and one element of the second matrix corresponds to the reflection characteristic value corresponding to one grid point; the output of the wireless signal loss prediction value of each grid point in the indoor space includes: outputting a third matrix, where one element of the third matrix corresponds to the wireless signal loss prediction value of one grid point in the indoor space.

[0013] Secondly, embodiments of this application provide a wireless signal loss prediction device, comprising: an acquisition module and a processing module; the acquisition module is configured to acquire the cumulative penetration loss value of a first path corresponding to each grid point in an indoor space, wherein the first path corresponding to each grid point is a path from the grid point to a target signal transmitter; the cumulative penetration loss value of a first path is the sum of the penetration loss values ​​of the wireless signal emitted by the target signal transmitter through each obstacle on the first path; the acquisition module is configured to acquire the reflection characteristic value corresponding to each grid point; the reflection characteristic value is determined based on the direct path loss value of the wireless signal to each grid point and the minimum reflection path loss value of the wireless signal to each grid point; and based on the cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point, the loss value of the wireless signal in the indoor space is predicted.

[0014] One possible implementation is that the acquisition module is specifically used to: determine the penetration loss value corresponding to each obstacle in the first path traversed by the first grid point; the first grid point is one of the grid points; and perform cumulative calculation on the penetration loss value corresponding to each obstacle to determine the cumulative penetration loss value corresponding to the first path where the first grid point is located.

[0015] In another possible implementation, before the acquisition module acquires the reflection feature value corresponding to each grid point, the acquisition module is further used to acquire the first free space propagation loss value of the wireless signal to the first grid point; the first grid point is one of the grid points; the processing module is further used to add the cumulative penetration loss value corresponding to the first path where the first grid point is located to the first free space propagation loss value to obtain the direct path loss value of the wireless signal to the first grid point.

[0016] In another possible implementation, before the acquisition module acquires the reflection feature value corresponding to each grid point, the acquisition module is further configured to acquire the second free space propagation loss value of the wireless signal to the first reflection point; the first reflection point is one of the reflection points corresponding to the first grid point, and the first grid point is one of the grid points; the processing module is further configured to add the second free space propagation loss value, the cumulative penetration loss value corresponding to the second path where the first reflection point is located, and the reflection loss generated by the wireless signal to the first reflection point to obtain the reflection path loss value of the first grid point; wherein, the second path where the first reflection point is located is the path between the first reflection point and the target signal transmitter, and the cumulative penetration loss value corresponding to a second path is the sum of the penetration loss values ​​of the wireless signal emitted by the target signal transmitter through each obstacle on the second path.

[0017] Another possible implementation is that the above processing module is specifically used to: input the cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point into the wireless signal loss prediction model; and output the wireless signal loss prediction value of each grid point in the indoor space.

[0018] In another possible implementation, the processing module is specifically used to input the first matrix and the second matrix into the wireless signal loss prediction model; wherein, the first matrix is ​​constructed based on the position of each grid point in the indoor space and the cumulative penetration loss value of the first path corresponding to each grid point, and one matrix element of the first matrix corresponds to the cumulative penetration loss value of the first path corresponding to one grid point; the second matrix is ​​constructed based on the position of each grid point in the indoor space and the reflection feature value corresponding to each grid point, and one matrix element of the second matrix corresponds to the reflection feature value corresponding to one grid point; the processing module is specifically used to output a third matrix, and one matrix element of the third matrix corresponds to the predicted wireless signal loss value of one grid point in the indoor space.

[0019] Thirdly, this application provides an electronic device comprising: a processor and a memory; the memory stores a program or instructions executable on the processor, wherein the program or instructions, when executed by the processor, implement the method of the first aspect described above.

[0020] Fourthly, this application provides a readable storage medium on which a program or instructions are stored, which, when executed by a computer, implement the method of the first aspect described above.

[0021] Fifthly, this application provides a computer program product stored in a storage medium, which, when executed by a computer, implements the method described in the first aspect.

[0022] In a sixth aspect, embodiments of this application provide a chip including a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the method described in the first aspect.

[0023] The beneficial effects of the second to sixth aspects mentioned above are described in the corresponding description of the first aspect and will not be repeated here. Attached Figure Description

[0024] Figure 1 A schematic diagram of the network architecture for the application of a wireless signal loss prediction method provided in this application embodiment;

[0025] Figure 2 A flowchart illustrating a wireless signal loss prediction method provided in an embodiment of this application;

[0026] Figure 3 A flowchart illustrating a wireless signal loss prediction method provided in an embodiment of this application;

[0027] Figure 4 A flowchart illustrating a wireless signal loss prediction method provided in an embodiment of this application;

[0028] Figure 5 A flowchart illustrating another wireless signal loss prediction method provided in an embodiment of this application;

[0029] Figure 6 A flowchart illustrating another wireless signal loss prediction method provided in an embodiment of this application;

[0030] Figure 7 A flowchart illustrating another wireless signal loss prediction method provided in an embodiment of this application;

[0031] Figure 8 A schematic diagram of a process for predicting indoor wireless propagation loss using a traditional intelligent model is provided for an embodiment of this application;

[0032] Figure 9 This is a schematic diagram of a traditional intelligent model structure provided in an embodiment of this application;

[0033] Figure 10 A schematic diagram illustrating a process for predicting indoor wireless propagation loss using an expert knowledge-enhanced intelligent model, provided in this application embodiment;

[0034] Figure 11 A schematic diagram of an input tensor provided in an embodiment of this application;

[0035] Figure 12 A schematic diagram of an input tensor provided in an embodiment of this application;

[0036] Figure 13 A schematic diagram of a target output tensor provided in an embodiment of this application;

[0037] Figure 14 A schematic diagram illustrating the distribution of true wireless propagation loss values ​​provided in an embodiment of this application;

[0038] Figure 15 A schematic diagram of the structure of a wireless signal loss prediction device provided in an embodiment of this application;

[0039] Figure 16 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0040] The wireless signal loss prediction method, apparatus, equipment, medium, and program products provided in this application will now be described in detail with reference to the accompanying drawings.

[0041] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0042] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0043] The terms "at least one," "at least one of," etc., used in the specification and claims of this application refer to any one, any two, or a combination of two or more of the included items. For example, at least one of a, b, and c can mean: "a," "b," "c," "a and b," "a and c," "b and c," and "a, b, and c," where a, b, and c can be single or multiple. Similarly, "at least two" refers to two or more items, and its meaning is similar to that of "at least one."

[0044] In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0045] This application provides a wireless signal loss prediction method, apparatus, device, medium, and program product, which can be applied to scenarios of indoor wireless channel modeling and wireless signal propagation loss prediction. Examples include: mobile communication network planning and rapid deployment in 6G technology; indoor wireless communication digital twin construction and intelligent operation and maintenance; wireless coverage and interference optimization in smart buildings; communication performance simulation and verification in high-density complex indoor scenarios; 6G integrated sensing indoor high-precision positioning and perception; wireless communication and IoT deployment in industrial indoor scenarios; adaptive optimization of wireless communication in dynamically variable indoor scenarios; and standardized testing and evaluation of indoor channel performance of communication equipment.

[0046] With the development of 6G communication technology, high-precision and high-efficiency indoor wireless channel modeling has become a key challenge. Indoor scenarios are characterized by complex structures and diverse obstructions, leading to numerous reflections and transmissions in signal propagation, resulting in extremely high complexity in wireless signal propagation outcomes. Traditional indoor wireless channel modeling methods, such as those relying on ray tracing, are computationally expensive and cannot meet the rapid deployment requirements of systems like digital twins in the 6G context. Intelligent models based on neural networks have gained attention; therefore, there is an urgent need for an intelligent wireless signal propagation modeling and wireless signal propagation loss prediction method that can accurately extract potential wireless signal propagation-related features from indoor scenario data and adapt to the complex variations in electromagnetic wave propagation results. This would improve the accuracy, efficiency, and practicality of wireless channel modeling and wireless signal propagation loss prediction in indoor scenarios, supporting the high-performance design and optimization of 6G communication systems.

[0047] In the field of wireless communication, modeling and predicting wireless signal propagation is a widely concerned issue, especially path loss, which is the most fundamental feature needed to analyze coverage area, interference power, and other related indicators. In classic solutions, there are two main methods for modeling and predicting path loss:

[0048] Method 1: Relying on physical simulation calculations, this method is based on full-wave simulation and ray tracing techniques. Although it can achieve high accuracy, its computational cost increases exponentially with the complexity of the analysis environment. Furthermore, full-wave simulation and ray tracing are generally considered "one-off" methods, meaning that even minor environmental changes require a complete rerun of the simulation to ensure accurate representation of updated conditions.

[0049] Method 2: Relies on statistical models, which is also a modeling method frequently used in 3GPP technical documents. While simpler, these methods focus on the overall distribution of path loss in the scenario and lack the ability to describe the parameters of specific target radio links.

[0050] Because the aforementioned classic techniques struggle to balance accuracy and efficiency in large-scale prediction scenarios, those skilled in the art have proposed deep learning-based methods for predicting indoor propagation loss. These methods have gradually become a popular approach and can be broadly categorized as follows:

[0051] Method 3: The deep learning method based on Multilayer Perceptron (MLP) is the most classic deep learning approach. By constructing a deep MLP, information related to the wireless propagation process is used as input to the neural network. This includes engineering data such as transceiver locations and manually set wireless environment characteristic parameters. The neural network learns the correlation between input and output from the dataset, thereby modeling and predicting wireless signal propagation loss. However, while a sufficiently large MLP can theoretically achieve arbitrary precision in approximating the target mapping, in practice, due to the sparsity of the relevant engineering parameters, although it can achieve a certain level of generalization, it is difficult to accurately represent sufficient radio propagation information. This has long resulted in insufficient accuracy for this MLP-based method.

[0052] Method 4: Deep learning methods utilizing image information have attracted widespread attention in recent years. Neural networks, such as Convolutional Neural Networks (CNNs), can directly extract and analyze features from image information, thus avoiding the manual extraction of engineering parameters and preserving rich wireless propagation scene information. Especially after Residual Networks (ResNets) implemented backpropagation of convolutional networks of arbitrary depth, they can fully leverage the spatial correlation inference capabilities of CNNs, continuously improving the accuracy of path loss modeling and prediction by expanding the network model's dimensionality. However, because traditional CNNs like ResNets always expand tensors at the end and input them into an MLP to output the final result, this step carries the risk of losing spatial correlation information.

[0053] Method 5: Deep learning methods based on fully convolutional neural networks (WCNNs) have become one of the most popular approaches. They eliminate the reliance on microprocessor-based algorithms (MLPs) found in traditional CNNs, effectively preserving the spatial correlation information learned by the neural network for deeper reasoning. Since wireless signal propagation results always exhibit high correlation at receivers in adjacent locations, this design approach is highly suitable for modeling and predicting wireless propagation results. The WCNN architecture, exemplified by U-Net, is a typical representative of this type of deep learning method. It first achieved the mapping from input information such as satellite imagery to path loss in urban environments, and further attracted widespread attention in research on wireless channel correlation technologies in indoor environments.

[0054] Specifically, among related technologies, there is a method for wireless channel modeling of indoor scenes based on LiDAR perception and reconstruction. The core technical process of this method involves automatically and accurately reconstructing an indoor 3D scene using LiDAR point cloud data, integrating electromagnetic simulation, and ultimately achieving high-fidelity channel characteristic simulation. The entire technology chain can be divided into four core stages: data processing, structure extraction, internal reconstruction, and electromagnetic modeling and simulation.

[0055] Step 1: Data Acquisition and Efficient Preprocessing

[0056] First, dense 3D point clouds of the indoor scene are acquired using LiDAR, covering the ceiling, floor, walls, doors, windows, and furniture surfaces. To address the issues of large raw data volume and noise, the system performs key preprocessing steps: removing irrelevant outdoor point clouds through 3D bounding box clipping; employing an outlier removal algorithm for filtering and noise reduction; and compressing the data size using a random downsampling algorithm, laying the foundation for subsequent processing.

[0057] Step 2: Robust extraction of scene structural elements

[0058] The system projects the preprocessed 3D point cloud onto a 2D plane to generate a point cloud density map. For the facade structures (mainly walls and columns), it innovatively integrates the Hough transform and the random sample consensus algorithm. Simultaneously, by analyzing the statistical histogram on the height axis, the height parameters of the ceiling and floor can be stably extracted. Finally, the algorithm can automatically fit the optimal connection method based on the residual point cloud between structural endpoints, reconstructing a complete and coherent 3D framework of the room.

[0059] Step 3: Intelligent Identification and Reconstruction of Internal Components

[0060] After extracting the main structures, the system projects the walls into 2D wall images and uses an advanced zero-shot image semantic segmentation model to automatically identify and segment openings such as doors and windows. Subsequently, all identified structural elements (walls, ceilings, floors) are removed from the original point cloud to obtain independent furniture point clouds. To achieve efficient and regularized furniture reconstruction, the method employs parametric template deformation technology. A template library is pre-established for common furniture (such as L-shaped workbenches) with multiple shape parameters (such as length, width, and height). The template parameters are adjusted using optimization algorithms to achieve an optimal fit with the remaining point cloud data, thereby generating a set of accurate 3D furniture model parameters.

[0061] Step 4: 3D Electromagnetic Modeling and Ray Tracing Simulation

[0062] Finally, all extracted parameters (structural dimensions, furniture models, opening locations) are integrated, and a triangular mesh surface model of the entire indoor scene is constructed using a triangulation algorithm. Each triangular element is assigned the corresponding International Telecommunication Union (ITU) standard electromagnetic parameters (such as dielectric constant and conductivity) of its actual material, forming a simulation model that can be used for electromagnetic calculations. After importing this model into professional ray tracing software and configuring system parameters such as transmitting and receiving antennas and frequencies, the propagation path of electromagnetic waves in complex environments can be simulated, and effects such as reflection and diffraction can be accurately calculated. Finally, key characteristics such as channel impulse response and path loss at the receiving point are output, and the simulation results show a high degree of agreement with measured data.

[0063] The summary of cutting-edge deep learning methods for wireless channel modeling and wireless signal propagation loss prediction reveals that while current deep learning methods can capture the overall distribution of indoor wireless propagation loss, they are significantly inadequate in learning complex radio propagation phenomena such as reflection and diffraction. This can lead to unexpected errors when using these deep learning methods for wireless resource scheduling. This limitation stems from the models' insufficient ability to learn the fundamental mechanisms of radio propagation, while the signal strength changes during indoor propagation exhibit extremely high complexity. On one hand, in the process of straight-line propagation, in addition to attenuation caused by free-space propagation loss, radio signals also experience varying degrees of rapid attenuation due to penetration loss. On the other hand, when radio signals encounter indoor obstacles and undergo reflection and diffraction, their original propagation direction changes, exacerbating the abrupt changes in propagation loss distribution under multipath effects. A key characteristic of indoor environments is the ubiquitous reflective surfaces such as walls and ceilings, making indoor radio propagation highly complex. This poses a significant challenge to deep learning methods for radio propagation prediction, as neural networks struggle to directly learn such complex mapping relationships from measured data, often accompanied by substantial computational costs: model size and training dataset size. Current technology struggles to escape this predicament. A significant drawback of this approach is its over-reliance on the data-driven underlying logic of deep learning and the arbitrary precision approximation capability of neural network theory, while neglecting the objective physical laws of wireless signal propagation itself. This leads to a data-driven "trap," failing to fully integrate deep learning methods with the field of radio, resulting in neural networks remaining in a "black box" state and unable to exhibit good generalization and interpretability.

[0064] To address the aforementioned technical problems: 1) excessively high computational costs of physical simulation; 2) lack of accurate modeling capability for specific target wireless links in statistical models; and 3) the inability of neural network-based deep learning schemes for wireless propagation modeling and prediction to effectively capture indoor radio propagation patterns, this application provides a wireless signal loss prediction method, apparatus, device, medium, and program product. Guided by expert knowledge in the field of radio propagation, it achieves a technical solution capable of effectively learning complex indoor propagation loss distribution patterns. Expert knowledge, namely the cumulative penetration loss value and reflection characteristic value of each signal propagation path, is injected into the wireless signal loss prediction model. This not only reduces excessive reliance on training datasets but also rapidly improves the prediction accuracy and generalization ability of intelligent models, creating a highly interpretable and accurate intelligent wireless channel modeling and prediction method for indoor scenarios in the future 6G era.

[0065] The following description, in conjunction with the accompanying drawings, details the wireless signal loss prediction method, apparatus, device, medium, and program products provided in the embodiments of this application.

[0066] Figure 1 The diagram illustrates a network architecture for applying a wireless signal loss prediction method according to an embodiment of this application. For example... Figure 1 As shown, the network architecture includes a wireless signal loss prediction device 101 and a terminal device 102. The wireless signal loss prediction device 101 and the terminal device 102 are interconnected.

[0067] In some embodiments, the wireless signal loss prediction device 101 may be a server, a computer, or a processor or processing unit within a server or computer. The server may be a single server or a server cluster comprising multiple servers. It should be noted that the embodiments of this application do not limit the specific device form of the wireless signal loss prediction device 101. Figure 1 The example shown is a wireless signal loss prediction device 101, which is a single server.

[0068] In some embodiments, the terminal device may be a mobile phone, tablet computer, laptop computer, handheld computer, in-vehicle electronic device, mobile internet device (MID), augmented reality (AR) / virtual reality (VR) device, robot, wearable device, personal computer (PC), ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc., and the embodiments of this application do not specifically limit it. Figure 1 The example shown is a mobile phone, with terminal device 102 as an example.

[0069] In some embodiments, the terminal device 102 may instruct the wireless signal loss prediction device 101 to perform wireless signal propagation loss prediction. Then, the wireless signal loss prediction device 101 obtains the cumulative penetration loss value of a first path corresponding to each grid point in the indoor space. The first path corresponding to each grid point is the path between the grid point and the target signal transmitter. The cumulative penetration loss value of a first path is the sum of the penetration loss values ​​of the wireless signal emitted by the target signal transmitter through each obstacle on the first path. The device also obtains the reflection characteristic value corresponding to each grid point. The reflection characteristic value is determined based on the direct path loss value of the wireless signal to each grid point and the minimum reflection path loss value of the wireless signal to each grid point. Further, the wireless signal loss prediction device 101 predicts the loss value of the wireless signal in the indoor space based on the cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point.

[0070] It should be noted that the network architecture described in the embodiments of this application is for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and does not constitute a limitation on the technical solutions provided in the embodiments of this application. As network architectures evolve, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0071] See Figure 2 This is a flowchart illustrating a wireless signal loss prediction method provided in an embodiment of this application. Figure 2 As shown, the wireless signal loss prediction method provided in this application embodiment can be implemented by the above-mentioned wireless signal loss prediction device, specifically including the following steps 201 to 203.

[0072] Step 201: The wireless signal loss prediction device obtains the cumulative penetration loss value of the first path corresponding to each grid point in the indoor space.

[0073] In some embodiments, the aforementioned indoor space can be any indoor space.

[0074] For example, the aforementioned indoor space can be a bedroom, living room, kitchen, shopping mall space, factory space, or supermarket space.

[0075] In some embodiments, the aforementioned indoor space is a gridded indoor space, meaning that multiple grid points of interest are set in the indoor space. A grid point can be represented as (x, y).

[0076] In some embodiments, the first path corresponding to each grid point is the path between the grid point and the target signal transmitter.

[0077] For example, the above path is a straight path from a grid point to the target signal transmitter.

[0078] For example, the first path described above can be represented by the following formula (1).

[0079] (1)

[0080] in, This indicates the path along this straight line. i grid points on the grid.

[0081] In some embodiments, the cumulative penetration loss value of the first path is the sum of the penetration loss values ​​of the wireless signal emitted by the target signal transmitter through each obstacle on the first path.

[0082] For example, the wireless signal loss prediction device accumulates the penetration loss values ​​corresponding to each obstacle along the straight propagation path of the wireless signal from the target signal transmitter to a grid point, i.e., the first path mentioned above, and finally obtains the cumulative penetration loss value of the first path where the grid point is located.

[0083] In some embodiments, combined with Figure 2 ,like Figure 3 As shown, step 201 above can be implemented through steps 201a and 201b.

[0084] Step 201a: The wireless signal loss prediction device determines the penetration loss value of each obstacle in the first path traversed by the first grid point.

[0085] In some embodiments, the first grid point is one of the grid points mentioned above.

[0086] In some embodiments, the penetration loss values ​​for obstacles of different materials are different.

[0087] For example, the penetration loss value of gypsum board / lightweight interior walls is typically about 3-6 dB, that of brick walls / concrete block walls is typically about 8-20 dB, that of reinforced concrete load-bearing walls is typically about 20-40 dB, that of glass windows is typically about 3-6 dB, and that of coated glass / energy-saving glass is typically about 20-40 dB.

[0088] For example, the penetration loss value corresponding to each obstacle can be denoted as T, and the penetration loss value corresponding to each obstacle on the first path can be denoted as... .

[0089] Step 201b: The wireless signal loss prediction device performs cumulative calculation on the penetration loss value corresponding to each obstacle to determine the cumulative penetration loss value corresponding to the first path where the first grid point is located.

[0090] For example, the wireless signal loss prediction device accumulates the penetration loss values ​​corresponding to each obstacle along the straight propagation path of the wireless signal from the target signal transmitter to the first grid point, i.e., the first path, and finally obtains the cumulative penetration loss value of the first path where the first grid point is located.

[0091] For example, the above-mentioned cumulative penetration loss value can be denoted as tensor CT.

[0092] For example, the value of the tensor CT at grid point (x, y) can be represented by the following formula (2).

[0093] (2)

[0094] It is understandable that the above cumulative penetration loss value can describe the degree of obstruction that the indoor environment may cause to the signal link through which wireless signals are propagated.

[0095] Thus, since the cumulative penetration loss value of a wireless signal along its propagation path is used to characterize the impact of obstacles on the wireless signal during propagation, the cumulative penetration loss value can be used to calculate the potential loss of the target signal link caused by obstacles in the indoor environment, thereby improving the accuracy of wireless signal loss prediction.

[0096] Step 202: The wireless signal loss prediction device obtains the reflection characteristic value corresponding to each of the above grid points.

[0097] In some embodiments, the aforementioned reflection characteristic values ​​are determined based on the direct path loss value of the wireless signal to each grid point and the minimum reflection path loss value of the wireless signal to each grid point.

[0098] For example, the above direct path loss value can characterize the propagation loss of the wireless signal on the direct path from the target signal transmitter to the grid point.

[0099] For example, the above reflection path loss value can characterize the propagation loss of the wireless signal on the reflection path from the target signal transmitter through the reflection point to the grid point.

[0100] For example, the wireless signal loss prediction device adds the direct path loss value of the wireless signal to a grid point to the minimum reflection path loss value of the wireless signal to a grid point to obtain the reflection characteristic value corresponding to that grid point. Further, following the same steps, the direct path loss value of the wireless signal to each grid point and the minimum reflection path loss value of the wireless signal to each grid point are added separately to obtain the reflection characteristic value corresponding to each grid point.

[0101] For example, the above reflection eigenvalues ​​can be denoted as tensor DRP.

[0102] For example, the reflection characteristic value DRP corresponding to each grid point (x, y) can be represented by the following formula (3).

[0103] (3)

[0104] in, Let be the direct path loss value of the wireless signal from the target signal transmitter to the grid point (x, y). This represents the minimum reflection path loss value for the wireless signal from the target signal transmitter to the grid point (x, y).

[0105] It is worth noting that each grid point has at least one reflection point, so the number of reflection path loss values ​​for each grid point is also at least one. That is, one reflection line of a grid point corresponds to one reflection path loss value. The reflection feature value in this embodiment is determined based on the minimum reflection path loss value, but it is not limited to this and can also be determined based on other reflection path loss values.

[0106] It is understandable that the above formula (3) is the formula for calculating the reflection characteristic value of a grid point, that is, the formula for calculating the reflection characteristic value DRP corresponding to the grid point (x, y).

[0107] It should be noted that the "wireless signal to grid point" mentioned in the embodiments of this application can be understood as the wireless signal being sent from the target signal transmitter to the grid point.

[0108] Step 203: The wireless signal loss prediction device predicts the loss value of the wireless signal in the indoor space based on the cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point.

[0109] In some embodiments, the wireless signal loss prediction device inputs the cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point into the wireless signal loss prediction model, and outputs the wireless signal loss prediction value of each grid point in the indoor space.

[0110] In some embodiments, combined with Figure 2 ,like Figure 4 As shown, step 203 above can be implemented through step 203a as follows.

[0111] Step 203a: The wireless signal loss prediction device inputs the cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point into the wireless signal loss prediction model, and outputs the wireless signal loss prediction value of each grid point in the indoor space.

[0112] For example, the above wireless signal loss prediction model is a neural network structure model.

[0113] For example, the wireless signal loss prediction model mentioned above can be a U-Shaped Network (U-Net) model, a U-Net using Stacked Dilated Convolutions (SDUNet) model, or a Swin Transformer U-Net (SwinUNet) model, etc., and this application does not impose any restrictions.

[0114] Among them, SDUNet, based on U-Net, uses stacked dilated convolutions to expand the receptive field of CNNs and has shown superiority in cutting-edge academic research on radio propagation prediction. SwinUNet, based on U-Net, combines the Transformer architecture, which is currently receiving widespread attention, and uses an attention mechanism to replace convolution operations.

[0115] For example, the wireless signal loss prediction device inputs the cumulative penetration loss value of the first path corresponding to each grid point and the reflection feature value corresponding to each grid point into the wireless signal loss prediction model. After the encoder and decoder parts of the wireless signal loss prediction model, the device performs feature extraction and reconstruction on the input cumulative penetration loss value of the first path corresponding to each grid point and the reflection feature value corresponding to each grid point, and finally outputs the wireless signal loss prediction value of each grid point in the indoor space.

[0116] For example, in the encoder part, each layer ends with a max pooling operation to perform downsampling with a kernel size of 2×2 and a stride of 2. This can effectively reduce the spatial resolution of the feature map by half, compress the spatial dimension, extract feature information for deeper depth analysis, utilize more channel dimensions, and achieve more complex abstract semantic information reasoning.

[0117] For example, in the decoder section, each layer ends with a transposed convolution operation to perform upsampling with a kernel size of 2×2 and a stride of 2, using a zero-padding strategy to double the spatial dimensions to reconstruct the data distribution and restore it to a structure of the same size as the input tensor.

[0118] In one example, in addition to inputting the cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point into the wireless signal loss prediction model, the wireless signal loss prediction device also inputs the data tensors such as indoor planar structure image and spatial distance into the wireless signal loss prediction model after scaling them to a uniform size, so as to provide more prediction information for the prediction process and thus improve the accuracy of the prediction.

[0119] It should be noted that the method of inputting the cumulative penetration loss value and reflection characteristic value constructed based on expert knowledge features into the wireless signal loss prediction model proposed in this application has strong model versatility. It does not depend on a specific network structure and can significantly improve model performance by injecting physical prior information.

[0120] Thus, by adding tensors to the input wireless signal loss prediction model, namely the cumulative penetration loss and reflection characteristic values ​​of each signal propagation path, the prediction accuracy and generalization ability of the model can be improved, thereby increasing the accuracy of wireless signal loss prediction.

[0121] In some embodiments, combined with Figure 2 ,like Figure 5 As shown, step 203a above can be implemented through the following step 203a1.

[0122] Step 203a1: The wireless signal loss prediction device inputs the first matrix and the second matrix into the wireless signal loss prediction model and outputs the third matrix.

[0123] In some embodiments, the first matrix is ​​constructed based on the location of each grid point in the indoor space and the cumulative penetration loss value of the first path corresponding to each grid point.

[0124] In some embodiments, a matrix element of the first matrix corresponds to the cumulative penetration loss value of the first path corresponding to a grid point.

[0125] In some embodiments, the second matrix is ​​constructed based on the position of each grid point in the indoor space and the reflection characteristic value corresponding to each grid point.

[0126] In some embodiments, one matrix element of the second matrix corresponds to the reflection feature value of a grid point.

[0127] In some embodiments, one matrix element of the third matrix corresponds to the predicted wireless signal loss value of a grid point in the indoor space.

[0128] For example, the wireless signal loss prediction device generates a matrix based on the distribution of the cumulative penetration loss value corresponding to each grid point in the indoor space, and uses this matrix as input to the wireless signal loss prediction model. Similarly, it generates a matrix based on the distribution of the reflection characteristic value corresponding to each grid point in the indoor space, and uses this matrix as input to the wireless signal loss prediction model. The wireless signal loss prediction model then outputs a matrix based on the distribution of grid points in the indoor space, representing the predicted value for each grid point.

[0129] In this way, the wireless signal loss prediction device generates an input tensor matrix according to the structure of the indoor space, i.e. the arrangement of network points. This allows the wireless signal loss prediction model to combine the known indoor space structure during the calculation process, improving the versatility of the aforementioned cumulative penetration loss value and reflection characteristic value. As a result, the aforementioned cumulative penetration loss value and reflection characteristic value can also be used in prediction models with other model structures.

[0130] It should be noted that the execution order of steps 201 and 202 described above is not limited in this embodiment. For example, step 201 can be executed first, followed by step 202; or step 202 can be executed first, followed by step 201; or steps 201 and 202 can be executed simultaneously. Figure 2 This example illustrates the process of executing step 201 first, followed by step 202.

[0131] The wireless signal loss prediction method provided in this application fully considers the penetration loss value generated by the wireless signal penetrating obstacles in the indoor space and the reflection loss value generated by the wireless signal reflecting in the indoor space. This allows the wireless signal loss during the wireless signal propagation process to be predicted through multiple loss values, thereby improving the accuracy of predicting the wireless signal loss during the wireless signal propagation process.

[0132] In some embodiments, combined with Figure 2 ,like Figure 6 As shown, prior to step 202 above, the wireless signal loss prediction method provided in this application embodiment may further include the following steps 301 and 302.

[0133] Step 301: The wireless signal loss prediction device obtains the first free space propagation loss value of the wireless signal to the first grid point.

[0134] In some embodiments, the first grid point is one of the grid points mentioned above.

[0135] In some embodiments, the free space propagation loss value described above characterizes the signal power attenuation caused only by energy diffusion across a spherical surface as the wireless signal propagates in ideal free space.

[0136] For example, the first free space propagation loss value mentioned above is represented by the following formula (4).

[0137] (4)

[0138] in, It is the first grid point Distance to the target signal transmitter It is a sufficiently small positive number. It's frequency. It's the speed of light.

[0139] Step 302: The wireless signal loss prediction device adds the cumulative penetration loss value corresponding to the first path where the first grid point is located to the first free space propagation loss value to obtain the direct path loss value of the wireless signal to the first grid point.

[0140] For example, the direct path loss value of the first grid point can be represented by the following formula (5).

[0141] (5)

[0142] It is understandable that the wireless signal loss prediction device can obtain the direct path loss value of the wireless signal to each grid point by following the steps described above for obtaining the direct path loss value of the wireless signal to the first grid point.

[0143] Thus, the direct path loss value of the wireless signal after its direct path to each grid point can be calculated using the methods and formulas provided above. This allows the loss of the wireless signal during direct path to be fully considered in the prediction of wireless signal loss, thereby improving the accuracy of predicting wireless signal loss during the propagation process.

[0144] In some embodiments, combined with Figure 2 ,like Figure 7 As shown, prior to step 202 above, the wireless signal loss prediction method provided in this application embodiment may further include the following steps 401 and 402.

[0145] Step 401: The wireless signal loss prediction device obtains the second free space propagation loss value of the wireless signal to the first reflection point.

[0146] In some embodiments, the first grid point is one of the grid points mentioned above.

[0147] In some embodiments, the first reflection point is one of the reflection points corresponding to the first grid point.

[0148] For example, the reflection point corresponding to the grid point is the location where the wireless signal is reflected after propagating to the vicinity of the grid point.

[0149] It is understandable that a grid point can correspond to one or more reflection points, meaning that wireless signals near the same grid point can be reflected by one or more reflection points.

[0150] In some embodiments, the free space propagation loss value described above characterizes the signal power attenuation caused only by energy diffusion across a spherical surface as the wireless signal propagates in ideal free space.

[0151] For example, the second free space propagation loss value is represented by the following formula (6).

[0152] (6)

[0153] in, It is the first grid point The distance from the reflection point to the target signal transmitter is the sum of the distance from the target transmitter to the reflection point and the distance from the reflection point to the first network point. distance, It is a sufficiently small positive number. It's frequency. It's the speed of light.

[0154] Step 402: The wireless signal loss prediction device adds the second free space propagation loss value, the cumulative penetration loss value corresponding to the second path where the first reflection point is located, and the reflection loss generated by the wireless signal to the first reflection point to obtain the reflection path loss value of the wireless signal to the first grid point.

[0155] In some embodiments, the second path where the first reflection point is located is the path between the first reflection point and the target signal transmitter.

[0156] In some embodiments, the cumulative penetration loss value corresponding to the second path is the sum of the penetration loss values ​​of the wireless signal emitted by the target signal transmitter through each obstacle on the second path.

[0157] For example, the direct path loss value of the first grid point can be represented by the following formula (7).

[0158] (7)

[0159] in, This indicates the distance of the wireless signal from the target signal transmitter to the first reflection point during the reflection process. The resulting reflection loss. It is based on the wireless signal from the target signal transmitter through the first reflection point To grid point The distance of the reflection path is calculated. The wireless signal travels from the target signal transmitter through the first reflection point To grid point The cumulative loss value of penetration along the reflection path.

[0160] For example, the wireless signal loss prediction device calculates the reflection loss generated by the wireless signal to each reflection point corresponding to the first grid point, calculates the second free space propagation loss value of each reflection point corresponding to the first grid point, and calculates the cumulative penetration loss value corresponding to the second path of each reflection point corresponding to the first grid point. Then, according to the above process, the three are added together to calculate at least one reflection path loss value of the first grid point.

[0161] It is understandable that the wireless signal loss prediction device can obtain the direct path loss value of the wireless signal to each grid point by following the steps described above for obtaining the reflection path loss value of the wireless signal to the first grid point.

[0162] Thus, the reflection path loss value of the wireless signal to each grid point can be calculated using the methods and formulas provided above. Then, the minimum value of the reflection path loss value of each grid point is taken as the main reflection path loss value of the wireless signal to that grid point. This allows the loss during wireless signal reflection to be fully considered in the process of predicting wireless signal loss, thereby improving the accuracy of predicting wireless signal loss during the propagation process.

[0163] The wireless signal loss prediction method of this application will be described below through specific embodiments.

[0164] In the following embodiments, we first take the traditional, data-driven deep learning problem of indoor wireless propagation loss prediction as a foundation, using the U-Net model as a typical representative of wireless signal loss prediction models, and serve as the baseline model for using intelligent models to predict wireless signal propagation loss. We then briefly introduce the wireless signal loss prediction method of this application. Next, we introduce a knowledge-data dual-driven deep learning method as the core technical idea of ​​this application's indoor wireless channel modeling and wireless signal loss prediction method, which uses expert knowledge—specifically, the aforementioned cumulative penetration loss value and reflection characteristic value—to enhance the intelligent model. Then, we introduce how to construct an input tensor containing expert knowledge—namely, the aforementioned cumulative penetration loss value and reflection characteristic value—under the guidance of radio propagation theory. This serves as a key breakthrough in realizing the fusion of expert knowledge and deep learning. We explain how to implement the indoor wireless channel modeling and prediction method using expert knowledge to enhance the intelligent model, and provide relevant performance analysis.

[0165] For example, the following is an introduction to the deep learning method framework for wireless channel modeling and prediction in indoor scenarios, namely the aforementioned wireless signal loss prediction model:

[0166] Traditional, data-driven deep learning methods for predicting indoor wireless signal loss, such as Figure 8 As shown, by training with a large dataset, the wireless signal loss prediction model can fit the distribution of wireless signal loss during propagation with a sufficiently small error, realizing an intelligent model-based method for indoor wireless channel modeling and prediction.

[0167] For computational efficiency, the wireless propagation loss prediction problem has been constructed as an image regression problem, and deep learning methods using images as input have become the dominant approach in recent research. The input to the intelligent model is a tensor form of various information, such as a planar structural diagram of the indoor wireless network deployment environment. While different research works often exhibit unique designs, the commonly used input is a planar structural diagram reflecting the location and material information of obstacles such as indoor walls. The target output is the propagation loss generated by the target signal transmitter within the target area. Currently, CNNs, represented by U-Net and following an encoder-decoder architecture, have attracted widespread attention in indoor radio propagation prediction. They can effectively preserve spatial correlation information during inference, identify abrupt changes based on geometric location information, and handle complex changes during wireless signal propagation, making them an important baseline neural network structure. Therefore, this technical solution uses U-Net as a typical representative of intelligent wireless signal loss prediction models and provides a brief introduction.

[0168] like Figure 9As shown, the overall structure of the wireless signal loss prediction model follows an encoder-decoder paradigm. The left half, which directly receives the input, is the encoder, and the right half, which outputs the final result, is the decoder. Feature extraction and reconstruction are performed in each half, and the final output is a 1-dimensional tensor, representing the wireless propagation loss information at each location on the indoor plane. In the following explanation, the tensor will be described using the form of tensor height × tensor width × tensor dimension. For example, the final output of the wireless signal loss prediction model is a 512 × 512 × 1 tensor.

[0169] Regarding the input to the model, the data tensors such as indoor planar structure images and spatial distances are scaled down to a uniform size. In addition to these basic input information, additional information can also be used as tensor input to assist the learning and reasoning process of the intelligent model. We will not discuss this in detail here, but denote the input as a tensor of 512×512×C.

[0170] The convolutional kernels in the wireless signal loss prediction model are all 3×3, with a stride of 1 and a dilation rate of 1. To maintain training efficiency and model stability, a two-dimensional batch normalization layer (Batch Norm) is always followed after the convolution to normalize the tensor data, ensuring it remains in a distribution with a mean of 0 and a variance of 1. This, along with the ReLU activation function, forms a convolutional block.

[0171] In the encoder part, each layer ends with a max pooling operation to perform downsampling. The kernel size is 2×2 and the stride is 2. This can effectively reduce the spatial resolution of the feature map by half, compress the spatial dimension, extract feature information for deeper depth analysis, and utilize more channel dimensions to achieve more complex abstract semantic information reasoning.

[0172] In the decoder section, each layer ends with a transposed convolution (ConvTranspose) operation to perform upsampling. The kernel size is 2×2, the stride is 2, and a zero-padding strategy is used to double the spatial dimensions to reconstruct the data distribution and restore it to a structure with the same size as the input tensor.

[0173] For example, the encoder in the wireless signal loss prediction model consists of five layers, each containing two consecutive convolutions followed by a max pooling operation. As depth increases, the spatial resolution of the tensor decreases while the number of channels increases, gradually transforming low-level appearance features into higher-level abstract semantic representations. The decoder also consists of five layers, each beginning with a ConvTranspose operation followed by two consecutive convolutional blocks to reconstruct the data layout. In each layer, the tensor is copied before the Max Pooling operation in the encoder. The copied tensor is then concatenated with the upsampled result of ConvTranspose along the channel dimension, and the combined tensor is passed to subsequent convolutional blocks. This approach integrates deep and shallow semantic information, facilitating more comprehensive analysis. This design is similar to residual connections, enhancing gradient propagation, mitigating the vanishing or exploding gradient problem, and reducing the learning burden on deeper encoder-decoder layers.

[0174] Finally, in the output layer of the decoder in the wireless signal loss prediction model, a single convolution is used as the output head, with a kernel size of 1×1 and a stride of 1. The tensor information is integrated into a 512×512×1 tensor to generate the final result, which is the predicted wireless signal loss value for each grid point in the aforementioned indoor space.

[0175] The following is a framework for an indoor wireless channel modeling and prediction method that utilizes expert knowledge to obtain novel input tensor constructions, namely the aforementioned cumulative penetration loss value and the aforementioned reflection characteristic value, and inputs them into the wireless signal loss model:

[0176] Since the propagation of wireless signals is the result of the combined effects of various radio propagation phenomena such as transmission and reflection, and intelligent models are trained to fit the results of this combined effect, it is difficult to understand the underlying radio propagation mechanism. As a result, the generalization ability of this deep learning-based solution is significantly limited by the representativeness of the training dataset, making it difficult to guarantee accuracy in new wireless propagation loss prediction tasks.

[0177] The technical solution provided in this application embodiment suggests that, under the guidance of expert knowledge, novel inputs should be constructed in the form of a matrix to adapt to intelligent models, thereby realizing a novel deep learning method for indoor scene wireless channel modeling and prediction with strong interpretability and high prediction accuracy.

[0178] The technical solution provided in this application proposes a method for indoor wireless channel modeling and prediction using an expert knowledge-enhanced intelligent model, such as... Figure 10As shown, this method focuses on modeling and predicting propagation loss in indoor wireless networks. To overcome the limitations of traditional data-driven deep learning methods, guided by radio propagation theory, a novel input tensor is constructed based on the geometric distribution and material parameters of the indoor environment. This tensor, along with environmental information, serves as the input to the intelligent model, explicitly injecting expert knowledge in the field of radio propagation into the deep learning process. This allows indoor wireless propagation loss prediction to escape the predicament of implicitly learning radio propagation laws, achieving high-precision prediction of propagation loss results, enhancing the interpretability of deep learning methods, and enabling wide application to various image regression-based intelligent models.

[0179] The above has introduced traditional data-driven deep learning and intelligent models for predicting wireless propagation loss. Next, we will explain how to construct a new type of input tensor under the guidance of wireless propagation theory, namely the above-mentioned cumulative penetration loss value and the above-mentioned reflection characteristic value, to inject expert knowledge into the intelligent model, improve the knowledge-data dual-driven deep learning design, and realize an indoor wireless channel modeling and prediction method that uses expert knowledge to enhance the intelligent model.

[0180] Specifically, the following two aspects will be illustrated by example:

[0181] 1. Basic Environmental Characteristics

[0182] The characteristics of a propagation channel are highly dependent on the surrounding environment. Although deep learning methods can extract abstract features through inductive bias and do not require the same level of environmental detail as ray tracing, indoor environments, characterized by ubiquitous walls and enclosures, make the relationship between radio propagation and environmental distribution more complex, especially for low-frequency signals, which may exhibit various phenomena such as penetration, reflection, and diffraction. Therefore, it is essential to use physical quantities characterizing the electromagnetic properties of objects as input.

[0183] Taking an indoor space resolution of 0.25m as an example, the transmission loss and reflection loss under normal incidence are expressed in decibels (dB), such as... Figure 11 In (a) tensor T and Figure 11 The tensor R in (b) is shown, where tensor T is the transmission loss at each grid point, i.e., the penetration loss value of each obstacle mentioned above. Tensor R is the reflection loss value at each grid point, where the value 0 indicates that there are no obstacles. To capture the positional relationship between the indoor space and the target signal transmitter, the distance from each grid point to the target signal transmitter is labeled as the input tensor D, as shown in Figure 1. Figure 11 As shown in (c) above, these three tensors provide key physical information related to the indoor environment.

[0184] 2. Input parameters constructed based on expert knowledge, namely the aforementioned cumulative penetration loss value and the aforementioned reflection characteristic value.

[0185] In the theoretical context of wireless signal propagation, expert knowledge provides deep learning models with structured information that is difficult for the models themselves to reason about autonomously. This is of great significance for improving the accuracy and generalization ability of deep learning model predictions. This technical solution designs two tensors infused with expert knowledge, incorporating knowledge of wireless propagation through penetration and reflection, thereby achieving knowledge-data dual-driven deep learning.

[0186] First, in a gridded indoor environment, for each grid point of interest... The straight line connected to the target signal transmitter is represented by the above formula (1). Then, obtain each The penetration loss value corresponding to each obstacle is Next, the calculation is performed to determine the distance the wireless signal travels from the target signal transmitter along the path described above, i.e., the first path, to the grid point. The accumulated penetration loss value over time is used to obtain the accumulated penetration loss value at that grid point, and the tensor is used to calculate this value. exist The numerical value can be obtained by referring to formula (2) above. Figure 12 The tensor shown in (a) .

[0187] Understandable, This is not a precise representation of the obstruction effect, but rather a relatively abstract description of the degree of obstruction that the indoor environment may cause to the target signal link. This imprecise propagation loss reference value is used as expert knowledge input, and an intelligent model derives a precise propagation loss result.

[0188] Besides penetration (transmission), reflection plays a crucial role in indoor radio propagation, especially in the sub-6 GHz band. Due to the multiple interactions between electromagnetic waves and surrounding obstacles, the reflection path significantly impacts the received signal power. Since multiple reflections often lead to severe signal attenuation, the path of a single reflection can be considered a potentially important reflection path. In the technical solution provided in this application, the single reflection path with minimum propagation loss is referred to as the main reflection path, and its reflection characteristic value corresponds to the expert knowledge tensor. .

[0189] For example, for target grid points tensor This indicates that it has already followed the direct path. The direct path loss value of the received wireless signal can be calculated using the above formula (5). The free space propagation loss value in formula (5) can be calculated using the above formula (4).

[0190] For example, for target grid points Assuming the target grid point By the reflection point The wireless signal is received on the main reflection path at the point of reflection, and its corresponding reflection path loss value can be calculated using the above formula (7). Among them, the free space propagation loss value in formula (7) can be calculated using the above formula (4).

[0191] Finally, among all potential reflection points, the reflection path loss value corresponding to the reflection point that produces the minimum reflection loss is usually selected, and the above formula (3) is used to calculate the reflection characteristic value of the grid point, i.e., as shown in the figure. Figure 12 The tensor shown in (b) .

[0192] It is understandable that, despite the above reflection eigenvalues, i.e., tensors The calculation of propagation loss did not take into account the transmission power, which will introduce some error. However, it must be emphasized that... This approach avoids complex calculations such as Maxwell's equations and does not obtain precise results through simulations similar to ray tracing. Instead, it aims to infuse deep learning models with expertise in radio reflection, thereby enhancing their ability to learn reflection-related propagation characteristics. During training, the intelligent model effectively combines knowledge-driven prior information with a data-driven learning process to improve prediction accuracy.

[0193] The following provides a method for computing tensors The overall process specifically includes steps 1 through 9:

[0194] Step 1: Initialize the reflection propagation feature tensor .

[0195] For example, based on the discretized grid size of the indoor environment, a reflection propagation feature corresponding to the grid size of the indoor environment is initialized, namely the aforementioned reflection feature value tensor. .

[0196] It should be noted that the reflection propagation characteristic tensor is usually used... The initial value is set to zero.

[0197] Step 2: Traverse each grid location in the indoor environment , that is, the grid points mentioned above.

[0198] For example, for each grid location in the indoor environment The subsequent reflection propagation loss calculation steps are executed sequentially to construct the reflection propagation characteristic values ​​at the corresponding locations. .

[0199] Step 3: Calculate the direct path propagation loss, i.e., the direct path loss value mentioned above.

[0200] For example, for the current grid position Based on propagation distance and penetration loss, calculate the direct path propagation loss of the wireless signal from the transmitter to this location along a straight path. At the same time, the interaction positions between the signal and the obstacle are recorded as potential reflection points.

[0201] Step 4: Initialize the reflection path propagation loss parameters.

[0202] For example, the reflection path propagation loss at the current grid position, i.e., the aforementioned reflection path loss value, is initialized to a preset maximum value for subsequent filtering of the main reflection path.

[0203] Step 5: Traverse the set of potential reflection points.

[0204] For example, for a plurality of identified potential reflection points, a reflection path loss estimation process is performed one by one, including: acquiring the wireless signal at the potential reflection points. Reflection loss generated at the location Wireless signals travel from the transmitter to the potential reflection point. —Target grid location Reflection propagation path distance, free space propagation loss And cumulatively calculate the penetration loss along the reflection path. This is the cumulative penetration loss value mentioned above. Based on the sum of these three values, the reflection path propagation loss of the corresponding candidate reflection path is evaluated. .

[0205] Step 6: Determine the propagation loss of the main reflection path.

[0206] For example, among all potential reflection paths, select The smallest reflection path is taken as the main reflection path at the current grid position, and the propagation loss corresponding to this path is taken as the reflection propagation loss value.

[0207] Step 7: Combine the propagation losses of the direct path and the reflection path.

[0208] For example, the direct path propagation loss obtained in step 3 is... Propagation loss of the dominant reflection path determined in step 6 By merging the values, the reflection propagation characteristic value at the current grid location is obtained. .

[0209] Step 8: Update the reflection propagation feature tensor .

[0210] For example, the above reflection propagation feature values ​​are stored in the reflection propagation feature tensor. The corresponding grid position.

[0211] Step 9: After traversing all grid locations, output the final constructed reflection propagation feature tensor. These are used as input features for subsequent intelligent models of wireless channel modeling.

[0212] Understandably, since the entire process does not involve complex computational logic, it is easier to implement GPU-based parallel accelerated computing, which is very friendly to actual deployment and application.

[0213] The technical proposal provided in this application has described how to construct an expert knowledge tensor under the guidance of wireless propagation theory in order to achieve the integration of data-driven and knowledge-driven processes in deep learning. The following is the overall process of the indoor wireless channel modeling and prediction method using an expert knowledge-enhanced intelligent model, specifically including the following steps A1 to A5.

[0214] Step A1: Construct a data-driven traditional indoor scene wireless channel prediction deep learning model, i.e., a wireless signal loss prediction model, based on the aforementioned "Indoor Scene Wireless Channel Modeling and Prediction Deep Learning Method Framework". The technical proposal provided in this application uses the U-Net model as an example to illustrate a general intelligent model, but it is not limited to the U-Net model.

[0215] Step A2: Organize the dataset used for model training. Discretize the entire indoor environment into a grid with a certain spatial resolution, such as 0.25m, forming a matrix format.

[0216] Step A3: Referring to the technical methods described above for constructing the cumulative penetration loss value and reflection characteristic value, construct an input tensor that incorporates expert knowledge.

[0217] Step A4: Train the initialized smart model using the training set data, employing backpropagation, using the root mean square error (RMSE) as the loss function, a batch size of 2, 200 training epochs, and an initial learning rate of... Cosine annealing is used to control convergence, with a lower bound on the learning rate of 1. .

[0218] Step A5: A well-trained intelligent model possesses the ability to predict wireless propagation loss in a target indoor scene. Input information for the target scene is constructed and normalized according to the normalization parameters set in the training set. This information is then input into the intelligent model, and the output is denormalized according to the training set settings to obtain the wireless propagation loss distribution in the target indoor scene. Specifically, as shown below... Figure 13 The diagram shows the target output tensor.

[0219] As can be seen from the above, the indoor wireless channel modeling and prediction method using an expert knowledge-enhanced intelligent model provided in this application embodiment addresses the shortcomings of the aforementioned related technologies. This invention proposes a knowledge-data dual-driven deep learning scheme. Guided by radio propagation theory, it constructs an input tensor infused with expert knowledge of radio propagation based on the geometric distribution information of the indoor environment and radio-related information. This user-friendly mode facilitates intelligent model retrieval, explicitly endowing the deep learning model with the laws of radio propagation. This overcomes the "black box" working mode of traditional deep learning, which directly fits the training dataset to approximate radio propagation phenomena, and overcomes shortcomings such as poor interpretability, high dependence on the training dataset, and insufficient generalization.

[0220] The embodiments provided in this application have illustrated how to implement an indoor wireless channel modeling and prediction method using an intelligent model enhanced by expert knowledge. The following section presents some practical test results.

[0221] The training and test sets can be taken from publicly available datasets, such as those with a spatial resolution of 0.25m, where the transmitter (signal sender) and receiver are set at a uniform height of 1.5m above the ground, using an omnidirectional antenna. The ray tracing simulation can handle up to 8 reflections, 10 transmissions, and 2 diffractions, closely resembling indoor radio propagation. Considering the limited dataset size, the data is divided into training and test sets in a 9:1 ratio, a common practice in related works. All data is normalized to maintain training stability and prevent information leakage from the test set.

[0222] To fully demonstrate the performance advantages of the expert knowledge enhancement scheme based on wireless propagation theory in the proposed technique, ablation experiments were conducted. Comparative analysis showed that as more expert knowledge is introduced during deep learning, the accuracy of the intelligent model's predictions is significantly improved. All models were implemented in PyTorch and trained and tested on a system equipped with an NVIDIA RTX 4090 GPU and an Intel(R) Xeon(R) Gold 6430 CPU.

[0223] As shown in Table 1 below, the performance of different intelligent models—specifically, wireless signal loss prediction models—in predicting indoor wireless propagation loss distribution was evaluated and compared using different input information configurations. Mean Absolute Error (MAE) was used as the metric to distinguish it from the loss function RMSE. In these experiments, the only change was the input tensor; the intelligent models with different architectures were trained and tested under the same settings. Whether in U-Net, SDUNet, or SwinUNet, compared to only inputting basic environmental information… The solution incorporates expert knowledge. The errors of the subsequent models all decreased significantly.

[0224] Specifically:

[0225] U-Net: The MAE on the test set decreased from 4.844dB to 2.268dB, and the error was reduced by approximately 53%.

[0226] SDUNet: MAE on the test set decreased from 3.875dB to 2.429dB.

[0227] SwinUNet: Test set MAE decreased from 4.488dB to 2.623dB

[0228] This demonstrates that a purely data-driven approach is insufficient to learn complex wireless propagation physics from limited data. The "expert knowledge enhancement" proposed in this application solves this problem by injecting wireless propagation physics mechanisms.

[0229]

[0230] Table 1

[0231] Further analysis of the data in Table 1 shows that the penetration accumulation feature constructed in the embodiments of this application... and main reflection path characteristics That is, both the aforementioned cumulative penetration loss value and the aforementioned reflection characteristic value have irreplaceable technical contributions. Firstly, adding [something] to the basic input... After tensor injection, the errors of each model decreased significantly, indicating that penetration loss is the main factor in indoor signal attenuation, and explicitly injecting this information is crucial. Secondly, further increasing... After tensor optimization, the test set errors of each model (especially SDUNet and SwinUNet) were further optimized (e.g., SDUNet decreased from 2.768dB to 2.429dB). This demonstrates that the proposed solution in this embodiment... Successfully enabling intelligent models to capture reflection components that significantly influence signal propagation, such as... Figure 14 This demonstrates the ability to more accurately predict complex signal intensity distributions. Among them, Figure 14 In the diagram, (a) represents the distribution of the true value of wireless propagation loss. Figure 14 In the diagram, (b) represents the predicted distribution when the inputs are T, R, D, and CT. Figure 14 In the equation (c), the predicted distribution is given by the inputs T, R, D, CT, and DRP.

[0232] Experimental results also verify the versatility of the technical solution provided in this application. As shown in Table 1, whether it is U-Net based on standard convolution, SDUNet based on dilated convolution, or SwinUNet combined with the Transformer architecture, all have achieved significant performance improvements after adopting the expert knowledge enhancement method proposed in this embodiment. This demonstrates that the core innovation of the technical solution provided in this application—"construction and injection of expert knowledge feature tensors"—has good applicability and model-agnostic characteristics, does not depend on a specific network structure, can be widely applied to various deep learning models, and has extremely high engineering application value and technological evolution potential.

[0233] The following is a detailed description of the key technical points and protection points of the technical solutions provided in the above embodiments:

[0234] The core of this embodiment lies in proposing an expert knowledge feature construction method based on the physical mechanism of wireless propagation and its application in indoor wireless channel prediction. By transforming the "penetration" and "reflection" mechanisms in wireless propagation into tensor data that can be directly processed by neural networks, a dual-driven approach of knowledge and data is achieved. Its specific innovations are reflected in:

[0235] 1. A model-independent general enhancement paradigm based on wireless propagation expert knowledge

[0236] Innovative Mechanism: Without performing complex ray tracing simulations, a simplified model of free-space propagation, penetration loss, and reflection loss is used to quickly generate an expert knowledge feature tensor that reflects the wireless propagation mechanism. This expert knowledge feature is represented in matrix / tensor form, allowing it to be directly used as input to a fully convolutional deep learning model. Furthermore, the constructed expert knowledge tensor (… It has high versatility and does not depend on a specific neural network structure.

[0237] Experimental Support: Experimental data shows that this feature construction method achieves significant performance improvements (MAE reduction of approximately 40%-53%) on standard convolutional networks (U-Net), dilated convolutional networks (SDUNet), and the Transformer architecture (SwinUNet). This demonstrates that the technical solution provided in this embodiment is a general enhancement paradigm with extremely high adaptability to technological evolution.

[0238] 2. Structural Construction Technique for Reflection Features Based on Simple Physical Calculations

[0239] This embodiment proposes a low-cost expert knowledge feature construction method for deep learning. Its core lies in replacing the complex electromagnetic simulation process with simplified geometric / physical approximation calculations, thereby constructing feature tensors with explicit physical meaning. (Guiding model training. Specific innovations include:)

[0240] Innovative Mechanism: Addressing the complex reflection phenomena indoors, this approach avoids relying on computationally expensive ray tracing algorithms for precise multipath search and complex Maxwell's equations. Instead, it discretizes the wireless propagation process onto a gridded space based on the geometric distribution, material parameters, and transmitter location within the indoor environment. Specifically, it introduces the concept of a "primary reflection path," estimating the propagation loss along the "transmitter-reflection point-receiver point" path and using only the primary reflection path with the minimum loss to represent the reflection characteristics at that location.

[0241] Technical advantages: It avoids the exponential computational complexity of traditional ray tracing simulations and achieves this through explicit tensor input ( This technique enables deep learning models to successfully capture reflection components that significantly influence signal strength. It allows the system to obtain prior information containing key physical laws without performing time-consuming simulations, achieving a dual optimization of computational efficiency and interpretability. Experiments have shown that this technique can significantly reduce the prediction error of intelligent models.

[0242] It should be noted that the above-mentioned technical methods are not precise calculations of wireless propagation results, but rather inject the wireless propagation mechanism into the deep learning process in the form of prior information, guiding the training and learning of the intelligent model, and significantly improving the model's learning and generalization capabilities for wireless propagation phenomena such as reflection and transmission.

[0243] 3. A collaborative mechanism driven by both knowledge and data.

[0244] This embodiment breaks away from the traditional deep learning approach that relies solely on data-driven methods. By explicitly injecting prior expert knowledge, it guides the model to learn the intrinsic laws of wireless propagation. The innovative value of this mechanism lies in: reducing the model's dependence on large-scale labeled datasets and decreasing the computational cost of training; enhancing the model's ability to learn complex indoor propagation phenomena, thereby improving the generalization and interpretability of prediction results; and achieving a balance between accuracy and efficiency, significantly reducing computational costs compared to physical simulation methods and significantly improving prediction accuracy compared to traditional data-driven models.

[0245] Based on the above key technical points, the protection points applied for in this embodiment regarding the knowledge-data dual-driven method that integrates expert knowledge and deep learning are as follows:

[0246] An explicit injection method for wireless propagation theory in deep learning: Based on the wireless propagation mechanism, construct a user-friendly input for the intelligent model and encode expert knowledge into the model input;

[0247] A method for constructing reflection characteristic tensors based on a simplified physical model: This method uses physical mechanism-based approximate calculations to replace full-wave simulations or ray-tracing simulations to characterize the indoor reflection environment. Important reflection paths are selected based on minimum loss, and the loss values ​​of these paths are mapped to tensors. This allows for the characterization of indoor reflective environments at a low computational cost, providing low-computational-cost physical prior guidance for deep learning models.

[0248] The following describes the feasibility of the wireless signal loss prediction method proposed in this application in terms of engineering implementation, system integration, and industrial application, specifically in the following aspects:

[0249] 1. The technical implementation path is clear, and the engineering feasibility is strong.

[0250] The technical solution provided in this embodiment is based on an expert knowledge-enhanced intelligent model, guided by wireless propagation theory.

[0251] The method for constructing distance, penetration and reflection features based on wireless propagation theory can be implemented on existing computing platforms through regularized computation, without relying on complex ray tracing simulation engines.

[0252] Training and inference can be completed using a standard deep learning framework;

[0253] Therefore, the technical solution provided in this embodiment can be fully implemented under existing computing conditions and software environment, and the difficulty of engineering implementation is low.

[0254] 2. It has moderate requirements for data and computing power, and is easy to deploy in practice.

[0255] Compared to traditional wireless propagation modeling methods that rely on high-precision 3D modeling and large-scale ray tracing simulation, this embodiment introduces expert knowledge features to explicitly inject prior information about the propagation mechanism during the model input stage:

[0256] The construction process of expert knowledge input involves only basic matrix operations, which facilitates efficient parallel computing and deployment;

[0257] It can enhance the accuracy and generalization ability of the model and effectively reduce the dependence on large-scale labeled data;

[0258] This reduces the need for massive computing resources during model training;

[0259] The model inference stage only requires feature calculation and neural network forward propagation, and the computational complexity is controllable.

[0260] This feature allows the technical solution provided in this embodiment to be deployed on cloud servers, local servers, or edge computing nodes, meeting the computing power requirements of different application scenarios.

[0261] 3. Clear application scenarios and urgent industry needs

[0262] The embodiments provided here are applicable to a variety of practical application scenarios, including but not limited to:

[0263] Indoor coverage modeling and optimization for 6G and future mobile communication systems;

[0264] Wireless channel prediction in complex indoor environments such as large buildings, parks, and shopping malls;

[0265] Modeling the propagation environment in a wireless network digital twin system;

[0266] The above-mentioned application scenarios all have a real need for high-precision, low-cost, and scalable wireless channel modeling technology. Therefore, the technical solution provided in this embodiment can effectively make up for the problems of insufficient accuracy of traditional statistical models and excessive cost of ray tracing methods, and has significant practical application value.

[0267] 4. Possesses smooth evolution capability

[0268] The knowledge-data dual-driven paradigm constructed using expert knowledge tensors provided in this embodiment has good scalability:

[0269] The technical solution provided in this embodiment has obvious model independence. As AI model architecture evolves (Transformer architecture and even future deep learning architecture), this input construction scheme can remain unchanged. Only the backend inference model needs to be updated to continuously obtain performance gains, thus possessing long-term technical vitality.

[0270] The features of expert knowledge can be expanded or adjusted according to different wireless frequency bands and different indoor scenarios;

[0271] The technical solution can be smoothly evolved to multi-band or three-dimensional spatial modeling scenarios.

[0272] Therefore, this embodiment is not only applicable to current wireless systems, but also has the ability to continuously evolve to future communication systems.

[0273] In summary, this embodiment has good conditions for implementation in terms of technical feasibility, engineering deployment feasibility, system integration feasibility, and industrial application value. It can be quickly transformed into actual products or system functions, and is suitable for promotion and application in the field of indoor wireless communication. It has significant engineering practical value and industrial promotion prospects.

[0274] It should be noted that the above-described method embodiments, or the various possible implementations of the method embodiments, can be executed individually, or, provided there is no conflict, they can be combined with each other. The specific implementation can be determined according to actual usage requirements, and this application embodiment does not impose any restrictions on this.

[0275] As can be seen, the above mainly describes the solutions provided by the embodiments of this application from a methodological perspective. To achieve the above functions, the embodiments of this application provide corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the modules and algorithm steps of the various examples described in the embodiments disclosed herein, the embodiments of this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0276] This application embodiment can divide the wireless signal loss prediction device into functional modules according to the above method example. For example, each function can be divided into its own functional module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware or as a software functional module. Optionally, the module division in this application embodiment is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0277] In some embodiments, this application also provides a wireless signal loss prediction apparatus. The wireless signal loss prediction apparatus may include one or more functional modules for implementing the wireless signal loss prediction method of the above method embodiments.

[0278] For example, Figure 15 This is a schematic diagram of a wireless signal loss prediction device provided in an embodiment of this application. Figure 15 As shown, the wireless signal loss prediction device 900 includes an acquisition module 901 and a processing module 902.

[0279] The aforementioned acquisition module 901 is used to acquire the cumulative penetration loss value of the first path corresponding to each grid point in the indoor space. The first path corresponding to each grid point is the path between the grid point and the target signal transmitter. The cumulative penetration loss value of a first path is the sum of the penetration loss values ​​of the wireless signal emitted by the target signal transmitter through each obstacle on the first path.

[0280] The aforementioned acquisition module 901 is used to acquire the reflection characteristic value corresponding to each grid point; the reflection characteristic value is determined based on the direct path loss value of the wireless signal to each grid point and the minimum reflection path loss value of the wireless signal to each grid point.

[0281] Based on the cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point, the loss value of the wireless signal in the indoor space is predicted.

[0282] In some embodiments, the acquisition module 901 is specifically used for:

[0283] Determine the penetration loss value of each obstacle in the first path traversed by the first grid point; the first grid point is one of the grid points mentioned above.

[0284] The cumulative penetration loss value corresponding to each obstacle is calculated to determine the cumulative penetration loss value corresponding to the first path where the first grid point is located.

[0285] In the wireless signal loss prediction device provided in this application embodiment, the wireless signal loss prediction device fully considers the penetration loss value generated by the wireless signal penetrating obstacles in the indoor space and the reflection loss value generated by the wireless signal reflecting in the indoor space. This allows the wireless signal loss during the wireless signal propagation process to be predicted through multiple loss values, thereby improving the accuracy of predicting the wireless signal loss during the wireless signal propagation process.

[0286] In some other embodiments, before the acquisition module 901 acquires the reflection feature value corresponding to each grid point, the acquisition module 901 is further configured to acquire the first free space propagation loss value of the wireless signal to the first grid point; the first grid point is one of the grid points in the above-mentioned grid points.

[0287] The aforementioned processing module 902 is further configured to add the cumulative penetration loss value corresponding to the first path where the first grid point is located to the first free space propagation loss value to obtain the direct path loss value of the wireless signal to the first grid point.

[0288] In some other embodiments, before the acquisition module 901 acquires the reflection feature value corresponding to each grid point, the acquisition module 901 is further configured to acquire the second free space propagation loss value of the wireless signal to the first reflection point; the first reflection point is one of the reflection points corresponding to the first grid point, and the first grid point is one of the grid points in each of the above grid points.

[0289] The processing module 902 is further configured to add the second free space propagation loss value, the cumulative penetration loss value corresponding to the second path where the first reflection point is located, and the reflection loss generated by the wireless signal to the first reflection point to obtain the reflection path loss value of the first grid point.

[0290] The second path where the first reflection point is located is the path between the first reflection point and the target signal transmitter. The cumulative penetration loss value corresponding to a second path is the sum of the penetration loss values ​​of the wireless signal emitted by the target signal transmitter through each obstacle on the second path.

[0291] In some other embodiments, the processing module 902 described above is specifically used for:

[0292] Input the cumulative penetration loss value of the first path corresponding to each of the above grid points and the reflection characteristic value corresponding to each of the above grid points into the wireless signal loss prediction model.

[0293] Output the predicted wireless signal loss value for each grid point in the above indoor space.

[0294] In some other embodiments, the processing module 902 is specifically used to input the first matrix and the second matrix into the wireless signal loss prediction model; wherein, the first matrix is ​​constructed based on the position of each grid point in the indoor space and the cumulative penetration loss value of the first path corresponding to each grid point, and one matrix element of the first matrix corresponds to the cumulative penetration loss value of the first path corresponding to one grid point; the second matrix is ​​constructed based on the position of each grid point in the indoor space and the reflection feature value corresponding to each grid point, and one matrix element of the second matrix corresponds to the reflection feature value corresponding to one grid point; the processing module 902 is specifically used to output a third matrix, and one matrix element of the third matrix corresponds to the wireless signal loss prediction value of one grid point in the indoor space.

[0295] It should be noted that the wireless signal loss prediction device can implement all the processes implemented in the above method embodiments and achieve the same beneficial effects. To avoid repetition, it will not be described again here.

[0296] In the case where the functions of the integrated modules described above are implemented in hardware, this application provides a possible structural schematic diagram of the electronic device involved in the above embodiments. For example... Figure 16 As shown, the electronic device 90 includes: a processor 92, a communication interface 93, and a bus 94. Optionally, the electronic device 90 may also include a memory 91.

[0297] Processor 92 may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 92 may be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 92 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0298] Communication interface 93 is used to connect with other devices via a communication network. This communication network can be Ethernet, wireless access network, wireless local area network (WLAN), etc.

[0299] The memory 91 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0300] As one possible implementation, the memory 91 can exist independently of the processor 92. The memory 91 can be connected to the processor 92 via a bus 94 and is used to store instructions or program code. When the processor 92 calls and executes the instructions or program code stored in the memory 91, it can implement the wireless signal loss prediction method provided in the embodiments of this application.

[0301] In another possible implementation, memory 91 can also be integrated with processor 92.

[0302] Bus 94 can be an Extended Industry Standard Architecture (EISA) bus, etc. Bus 94 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 16 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0303] Through the above description of the implementation methods, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the service calling device can be divided into different functional modules to complete all or part of the functions described above.

[0304] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described wireless signal loss prediction method embodiment and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0305] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0306] This application also provides a readable storage medium storing a program or instructions that, when executed by a computer, implement the wireless signal loss prediction method provided in the above embodiments. It is understood that all or part of the processes in the above method embodiments can be executed by computer instructions instructing related hardware; the readable storage medium can be any of the foregoing embodiments or memory; the readable storage medium can also be an external storage device of the service invocation device, such as a plug-in hard drive, SmartMedia Card (SMC), Secure Digital (SD) card, flash card, etc., equipped on the service invocation device. Further, the readable storage medium can include both internal storage units of the service invocation device and external storage devices. The readable storage medium is used to store the computer program and other programs and data required by the service invocation device. The readable storage medium can also be used to temporarily store data that has been output or will be output.

[0307] This application also provides a computer program product, which is stored in a storage medium and, when executed by a computer, implements the wireless signal loss prediction method provided in the above embodiments.

[0308] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0309] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0310] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for predicting wireless signal loss, characterized in that, include: Obtain the cumulative penetration loss value of the first path corresponding to each grid point in the indoor space, where the first path corresponding to each grid point is the path between the grid point and the target signal transmitter; The cumulative penetration loss value of a first path is the sum of the penetration loss values ​​of the wireless signal emitted by the target signal transmitter through each obstacle on the first path; Obtain the reflection characteristic value corresponding to each grid point; the reflection characteristic value is determined based on the direct path loss value of the wireless signal to each grid point and the minimum reflection path loss value of the wireless signal to each grid point; Based on the cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point, the loss value of the wireless signal in the indoor space is predicted.

2. The wireless signal loss prediction method according to claim 1, characterized in that, The step of obtaining the cumulative penetration loss value of the first path corresponding to each grid point in the indoor space includes: Determine the penetration loss value of each obstacle in the first path traversed by the first grid point; the first grid point is one of the grid points in the first grid point; The cumulative penetration loss value corresponding to each obstacle is calculated to determine the cumulative penetration loss value corresponding to the first path where the first grid point is located.

3. The wireless signal loss prediction method according to claim 1, characterized in that, Before obtaining the reflection feature value corresponding to each grid point, the method further includes: Obtain the first free-space propagation loss value of the wireless signal to the first grid point; the first grid point is one of the grid points in the set of grid points. The cumulative penetration loss value corresponding to the first path where the first grid point is located is added to the first free space propagation loss value to obtain the direct path loss value of the wireless signal to the first grid point.

4. The wireless signal loss prediction method according to claim 1, characterized in that, Before obtaining the reflection feature value corresponding to each grid point, the method further includes: Obtain the second free space propagation loss value of the wireless signal to the first reflection point; the first reflection point is one of the reflection points corresponding to the first grid point, and the first grid point is one of the grid points in each grid point; The reflection path loss value of the first grid point is obtained by adding the second free space propagation loss value, the cumulative penetration loss value corresponding to the second path where the first reflection point is located, and the reflection loss generated by the wireless signal to the first reflection point. Wherein, the second path where the first reflection point is located is the path between the first reflection point and the target signal transmitter, and the cumulative penetration loss value corresponding to a second path is the sum of the penetration loss values ​​of the wireless signal emitted by the target signal transmitter through each obstacle on the second path.

5. The wireless signal loss prediction method according to any one of claims 1 to 4, characterized in that, The method of predicting the loss value of the wireless signal in the indoor space based on the cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point includes: The cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point are input into the wireless signal loss prediction model. Output the predicted wireless signal loss value for each grid point in the indoor space.

6. The wireless signal loss prediction method according to claim 5, characterized in that, The step of inputting the cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point into the wireless signal loss prediction model includes: Input the first and second matrices into the wireless signal loss prediction model; The first matrix is ​​constructed based on the position of each grid point in the indoor space and the cumulative penetration loss value of the first path corresponding to each grid point, wherein one element of the first matrix corresponds to the cumulative penetration loss value of the first path corresponding to a grid point; the second matrix is ​​constructed based on the position of each grid point in the indoor space and the reflection characteristic value corresponding to each grid point, wherein one element of the second matrix corresponds to the reflection characteristic value corresponding to a grid point. The step of outputting the predicted wireless signal loss value for each grid point in the indoor space includes: Output a third matrix, where each element of the third matrix corresponds to the predicted wireless signal loss value of a grid point in the indoor space.

7. A wireless signal loss prediction device, characterized in that, include: Acquisition module and processing module; The acquisition module is used to acquire the cumulative penetration loss value of the first path corresponding to each grid point in the indoor space, and the first path corresponding to each grid point is the path between the grid point and the target signal transmitter; The cumulative penetration loss value of a first path is the sum of the penetration loss values ​​of the wireless signal emitted by the target signal transmitter through each obstacle on the first path; The acquisition module is used to acquire the reflection feature value corresponding to each grid point; The reflection characteristic value is determined based on the direct path loss value of the wireless signal to each grid point and the minimum reflection path loss value of the wireless signal to each grid point; Based on the cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point, the loss value of the wireless signal in the indoor space is predicted.

8. The wireless signal loss prediction device according to claim 7, characterized in that, The acquisition module is specifically used for: Determine the penetration loss value of each obstacle in the first path traversed by the first grid point; the first grid point is one of the grid points in the first grid point; The cumulative penetration loss value corresponding to each obstacle is calculated to determine the cumulative penetration loss value corresponding to the first path where the first grid point is located.

9. The wireless signal loss prediction device according to claim 7, characterized in that, Before the acquisition module acquires the reflection feature value corresponding to each grid point, the acquisition module is also used to acquire the first free space propagation loss value of the wireless signal to the first grid point; the first grid point is one of the grid points; The processing module is further configured to add the cumulative penetration loss value corresponding to the first path where the first grid point is located to the first free space propagation loss value to obtain the direct path loss value of the wireless signal to the first grid point.

10. The wireless signal loss prediction device according to claim 7, characterized in that, Before the acquisition module acquires the reflection feature value corresponding to each grid point, the acquisition module is also used to acquire the second free space propagation loss value of the wireless signal to the first reflection point; the first reflection point is one of the reflection points corresponding to the first grid point, and the first grid point is one of the grid points in each grid point; The processing module is further configured to add the second free space propagation loss value, the cumulative penetration loss value corresponding to the second path where the first reflection point is located, and the reflection loss generated by the wireless signal to the first reflection point to obtain the reflection path loss value of the first grid point. Wherein, the second path where the first reflection point is located is the path between the first reflection point and the target signal transmitter, and the cumulative penetration loss value corresponding to a second path is the sum of the penetration loss values ​​of the wireless signal emitted by the target signal transmitter through each obstacle on the second path.

11. The wireless signal loss prediction device according to any one of claims 7 to 10, characterized in that, The processing module is specifically used for: The cumulative penetration loss value of the first path corresponding to each grid point and the reflection characteristic value corresponding to each grid point are input into the wireless signal loss prediction model. Output the predicted wireless signal loss value for each grid point in the indoor space.

12. The wireless signal loss prediction device according to claim 11, characterized in that, The processing module is specifically used to input the first matrix and the second matrix into the wireless signal loss prediction model. The first matrix is ​​constructed based on the position of each grid point in the indoor space and the cumulative penetration loss value of the first path corresponding to each grid point, wherein one element of the first matrix corresponds to the cumulative penetration loss value of the first path corresponding to a grid point; the second matrix is ​​constructed based on the position of each grid point in the indoor space and the reflection characteristic value corresponding to each grid point, wherein one element of the second matrix corresponds to the reflection characteristic value corresponding to a grid point. The processing module is specifically used to output a third matrix, wherein one element of the third matrix corresponds to the predicted wireless signal loss value of a grid point in the indoor space.

13. An electronic device, characterized in that, It includes a processor and a memory, the memory storing a program or instructions that can run on the processor, the program or instructions being executed by the processor to implement the wireless signal loss prediction method as described in any one of claims 1 to 6.

14. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a computer, implement the wireless signal loss prediction method as described in any one of claims 1 to 6.

15. A computer program product, characterized in that, The computer program product is stored in a storage medium, and when executed by a computer, the computer program product implements the wireless signal loss prediction method as described in any one of claims 1 to 6.

Images