Indoor radio map construction method and device based on physical enhanced diffusion model

Abstract

The application discloses a kind of indoor radio map construction method and device based on physical enhanced diffusion model, comprising: according to the geometric information and building material information of indoor environment, the initial discontinuity physical prior including diffraction point set, transmission boundary point set is constructed;The diffraction point set in initial discontinuity physical prior is pruned to update discontinuity physical prior;According to indoor radio access point and updated discontinuity physical prior, generate discontinuity physical prior map;The multi-modal physical prior including the coordinate position of indoor radio access point, the reflection coefficient map and transmission coefficient map of the building material at each position, the discontinuity physical prior map is constructed;After multi-modal physical prior is processed by a multi-modal fusion encoder, the processing result is input as model input condition into the decoupled diffusion model trained, to generate indoor radio map.The application can quickly generate high-fidelity indoor radio map without field measurement.

Description

Technical Field

[0001] This invention belongs to the field of communication technology, specifically relating to a method and apparatus for constructing indoor radio maps based on a physical enhancement diffusion model. Background Technology

[0002] Indoor positioning plays a central role in smart buildings, industrial automation, emergency response, and immersive applications, with a growing demand for meter- to sub-meter-level positioning accuracy and scalable, low-latency deployment. Radio maps (RM), as an electromagnetic representation of the environment, correlate scene geometry, material properties, and the spatial distribution of signal strength, enabling high-precision positioning without costly on-site surveys. This makes RM a key technology for next-generation wireless communication and positioning services.

[0003] Currently, the technical solutions for constructing high-precision indoor radio maps mainly fall into three categories:

[0004] Ranging-based methods: These methods rely on measurements such as Received Signal Strength Indicator (RSSI) or Time of Arrival (ToA) and combine them with statistical distance loss models to estimate distance. For example, free-space path loss models or logarithmic distance path loss models are used to convert the received signal strength into distance from the transmitter, thereby enabling location.

[0005] Fingerprint-based methods: This method requires extensive offline surveying to densely collect signal fingerprints (such as RSSI, CSI, etc.) from a large number of location points within the target area, constructing a "location-fingerprint" database. During localization, the real-time measured signal fingerprints are matched with records in the database (e.g., using the K-nearest neighbor algorithm) to determine the user's location.

[0006] Simulation-based radio map construction methods aim to avoid large-scale field measurements by generating radio maps through software simulation. Specifically, these methods can be categorized as follows: 1) Full-wave electromagnetic solvers: These methods accurately simulate electromagnetic wave propagation by directly solving Maxwell's equations, theoretically offering the highest accuracy. 2) Ray tracing: An approximation method that simplifies electromagnetic waves into rays, simulating their reflection, transmission, and diffraction behavior in the environment to calculate signal strength distribution. 3) Neural network-based methods: These methods utilize deep learning models to directly learn the mapping relationship from environmental layout (e.g., floor plans), access point (AP) locations, to the radio map. For example, RadioUNet uses a U-Net network structure for end-to-end generation; RME-GAN employs a generative adversarial network (GAN); and SIP2Net introduces asymmetric convolution and dilated spatial pyramid pooling on top of U-Net to improve performance. These methods typically take environmental geometry as input and output pixel-level signal strength maps.

[0007] However, the aforementioned existing technologies have the following problems in practical applications:

[0008] Methods based on ranging models: Indoor environments are extremely complex, with severe multipath propagation, signal penetration, and diffraction effects. This causes the relationship between signal strength and distance to no longer satisfy a simple statistical model, resulting in huge ranging errors and making it impossible to guarantee positioning accuracy.

[0009] Fingerprint database-based methods: The construction and maintenance of fingerprint databases are extremely costly, requiring a significant amount of manpower and time for comprehensive signal collection. Furthermore, any minor changes in the indoor environment (such as moving furniture or opening / closing doors) can cause the fingerprint database to become invalid, necessitating frequent updates and exhibiting poor scalability.

[0010] Simulation-based methods have several drawbacks: 1) High computational latency: While full-wave electromagnetic solvers are accurate, their memory and computational requirements grow exponentially, limiting their application to small-scale simulations and making room or floor-level deployments impractical. Ray tracing, though faster, still requires minutes of computation time to handle complex multipath propagation, failing to meet the real-time update requirements of dynamic environmental changes. 2) Reliance on sparse measurements: Most existing neural network methods rely on some real, sparse signal measurement data for training or calibration, limiting their application in "zero-measurement" scenarios. 3) Poor physical consistency and insufficient generalization: Existing measurement-free neural network methods (such as RadioUNet) are mostly designed for outdoor scenarios, typically assuming homogeneous environmental materials. This assumption is completely invalid in indoor environments, which are filled with heterogeneous materials such as concrete, glass, and wood. This mismatch between model design and actual application conditions results in low accuracy of the generated radio maps at boundaries such as walls, doors, and windows, failing to accurately capture signal abrupt changes caused by material differences. Purely data-driven models struggle to learn material-sensitive propagation patterns, leading to poor generalization performance under new building layouts or material configurations. Summary of the Invention

[0011] To address the aforementioned problems in the existing technology, this invention provides a method and apparatus for constructing indoor radio maps based on a physical enhancement diffusion model. The technical problem to be solved by this invention is achieved through the following technical solution:

[0012] In a first aspect, embodiments of the present invention provide a method for constructing indoor radio maps based on a physical augmentation diffusion model, the method comprising:

[0013] Based on the geometric information of the indoor environment and the information of building materials, an initial discontinuous physical prior is constructed; the initial discontinuous physical prior includes the diffraction point set and the transmission boundary point set.

[0014] The diffraction point set in the initial discontinuous physical prior is pruned to update the discontinuous physical prior;

[0015] Generate a discontinuous physical prior map based on the indoor radio access point and the updated discontinuous physical prior.

[0016] Construct a multimodal physical prior; the multimodal physical prior includes the coordinate location of the indoor radio access point, the reflection coefficient map and transmission coefficient map characterizing the building materials at each location, and the discontinuous physical prior map;

[0017] The multimodal physical priors are processed by a multimodal fusion encoder, and the processing result is used as the model input condition to input into the trained decoupled diffusion model to generate an indoor radio map.

[0018] In one embodiment of the present invention, an initial discontinuity physical prior is constructed based on the geometric information of the indoor environment and the building material information, including:

[0019] Using a geometric neighborhood analysis algorithm, the effective corners where diffraction occurs are located based on the geometric information of the indoor environment, and the set of diffraction points is formed by all effective corners.

[0020] A transmission coefficient map is generated based on the building material information of the indoor environment. The transmission coefficient map is then thresholded to locate areas with strong transmission. A transmission boundary point set is formed by the boundary points of all areas with strong transmission.

[0021] In one embodiment of the present invention, pruning the diffraction point set in the initial discontinuous physical prior includes:

[0022] Based on the geometric information of the indoor environment, diffraction points that cannot form an effective incident shadow area in the diffraction point cluster are eliminated, as are diffraction points whose regions on both sides of the diffraction point cluster are geometrically connected and do not constitute a signal abrupt boundary.

[0023] In one embodiment of the present invention, the process of generating a discontinuous physical prior map includes:

[0024] Taking the indoor radio access point as the starting point, and the direction from the indoor radio access point to each diffraction point and transmission boundary point in the updated discontinuous physical prior as the extension direction, several rays are formed.

[0025] Remove the portion of each ray between the indoor radio access point and each diffraction point and transmission boundary point in the updated discontinuity physics prior, and generate a discontinuity physics prior map based on the remaining portion of each ray.

[0026] In one embodiment of the present invention, the multimodal fusion encoder is a convolutional layer with a 1×1 kernel.

[0027] In one embodiment of the present invention, the decoupling diffusion model includes a conditional coding module, a noise feature extraction module, and a dual-target recovery and fusion module; wherein...

[0028] The conditional encoding module is used to encode the model input conditions obtained by the multimodal fusion encoder through an independent conditional encoder to obtain conditional features;

[0029] The noise feature extraction module is used to extract noise features from the noisy indoor radio map of the current iteration using a U-Net encoder at multiple scales.

[0030] The dual-target recovery and fusion module includes two parallel U-Net decoders. Based on the two parallel U-Net decoders, noise features and conditional features are fused through a cross-attention mechanism. According to the fused features, the two parallel U-Net decoders recover two independent target outputs. One U-Net decoder recovers the prediction of the current noise component, and the other U-Net decoder recovers the direct prediction of the noise-free indoor radio map. Based on the two target outputs recovered by the dual-target recovery and fusion module, the noisy indoor radio map required for the next iteration is calculated and generated through a back-diffusion sampling algorithm until a clear indoor radio map is finally recovered.

[0031] In one embodiment of the present invention, the indoor radio map dataset used in the training process of the decoupled diffusion model includes an indoor radio map dataset generalized to antenna location and an indoor radio map dataset generalized to zero-sample layout.

[0032] Secondly, embodiments of the present invention provide an indoor radio map construction device based on a physical augmentation diffusion model, the indoor radio map construction device comprising:

[0033] The first construction module is used to construct the initial discontinuous physical prior based on the geometric information of the indoor environment and the building material information; the initial discontinuous physical prior includes the diffraction point set and the transmission boundary point set;

[0034] The pruning module is used to prune the diffraction point set in the initial discontinuous physical prior in order to update the discontinuous physical prior.

[0035] The first generation module is used to generate a discontinuous physical prior map based on the indoor radio access point and the updated discontinuous physical prior.

[0036] The second building module is used to construct multimodal physical priors; among which, multimodal physical priors include the coordinate location of the indoor radio access point, the reflection coefficient map and transmission coefficient map characterizing the building materials at each location, and the discontinuous physical prior map;

[0037] The second generation module is used to process the multimodal physical priors through a multimodal fusion encoder, and then use the processing results as input conditions to the trained decoupled diffusion model to generate an indoor radio map.

[0038] The beneficial effects of this invention are:

[0039] The indoor radio map construction method based on a physical enhancement diffusion model proposed in this invention overcomes the problems of high computational latency, reliance on field measurements, and insufficient modeling of complex indoor environments in existing indoor radio map construction technologies. It is a high-fidelity indoor radio map construction method that requires no field measurements, can generate maps quickly, and has the following significant advantages: By using the electromagnetic properties of materials and key physical structures (diffraction points, transmission boundary points) as priors, the indoor radio map generated by this invention can highly realistically reproduce the details of signal propagation in complex indoor environments, especially in boundary areas where signals change drastically, with accuracy far exceeding existing technologies; the method proposed in this invention requires no field measurements whatsoever. This invention generates high-precision indoor radio maps based solely on building transmission coefficient maps, reflection coefficient maps, individual access point (AP) locations, and discontinuous physical prior maps. This significantly reduces deployment costs and time, and supports rapid response to dynamic environmental changes. The invention employs a decoupled diffusion model. Because the model learns universal electromagnetic propagation physics rather than data distribution specific to a particular scenario, it can effectively generalize to entirely new indoor layouts and AP locations, exhibiting strong robustness. Experiments demonstrate that using the indoor radio map generated by this invention as a fingerprint database can achieve sub-10-meter indoor positioning accuracy, providing a solid technical foundation for various downstream applications requiring precise location information.

[0040] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating an indoor radio map construction method based on a physical enhancement diffusion model provided in an embodiment of the present invention.

[0042] Figure 2 This is a schematic diagram of generating a discontinuous physical prior graph provided in an embodiment of the present invention;

[0043] Figure 3 This is a schematic diagram of a network model adapted for indoor radio map generation provided in an embodiment of the present invention;

[0044] Figure 4 This is a schematic diagram of the iterative denoising process using a decoupling diffusion model provided in an embodiment of the present invention;

[0045] Figure 5 This is a comparative diagram of the visualization generation results of various methods in the antenna position generalization scenario provided by the embodiments of the present invention;

[0046] Figure 6 This is a comparative diagram of the visualization generation results of various methods in the zero-shot layout generalization scenario provided by the embodiments of the present invention;

[0047] Figure 7 This is a schematic diagram of an indoor radio map construction device based on a physical enhancement diffusion model provided in an embodiment of the present invention. Detailed Implementation

[0048] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0049] In order to overcome the shortcomings of existing radio map construction technologies, such as high computational latency, reliance on field measurements, and insufficient modeling of complex indoor environments, this invention aims to design a high-fidelity indoor radio map construction method that can be generated quickly without field measurements. To achieve this goal, the inventors discovered that the following issues need to be considered during the design process: First, the proposed method needs to adapt to the material heterogeneity of the indoor environment by explicitly incorporating the electromagnetic properties (such as reflection and transmission coefficients) of different materials such as walls and windows into the model, breaking the "homogeneous medium" assumption commonly found in traditional methods, thereby improving physical realism. Second, the proposed method must accurately model signal discontinuities, accurately capturing signal intensity abrupt changes caused by physical phenomena such as wall diffraction and door / window transmission, solving the problem that existing convolutional networks, due to their smooth priors, struggle to handle such non-stationary field distributions. Third, the proposed method should ensure physical consistency and high generalization ability, ensuring that the generated radio map not only conforms to the basic laws of electromagnetic wave propagation but can also effectively generalize to previously unseen indoor layouts and material configurations. Finally, the proposed method needs to support real-time updates, enabling rapid map regeneration by simply updating physical parameters when the environment changes, meeting the needs of dynamic scenarios. Ultimately, considering the above issues, this invention proposes a method and apparatus for constructing indoor radio maps based on a physically enhanced diffusion model.

[0050] Firstly, please see Figure 1 This invention provides a method for constructing indoor radio maps based on a physical augmentation diffusion model, specifically including the following steps:

[0051] S10. Based on the geometric information of the indoor environment and the information of building materials, construct the initial discontinuous physical priors; the initial discontinuous physical priors include the diffraction point set and the transmission boundary point set.

[0052] Unlike existing indoor radio map construction methods that only use geometric layout as input, this invention recognizes that non-stationary abrupt changes, or discontinuities, in indoor signals are the main source of prediction errors. These non-stationary abrupt changes are dominated by specific electromagnetic propagation phenomena such as diffraction and strong transmission, which standard convolutional networks, due to their inherent smooth priors, struggle to capture. Therefore, this invention proposes to explicitly encode these discontinuous priors as guiding information for the model. Specifically, in this embodiment, based on the geometric information of the indoor environment and building material information, an initial discontinuous physical prior is constructed, including: using a geometric neighborhood analysis algorithm to locate the effective corners where diffraction occurs based on the geometric information of the indoor environment, forming a diffraction point set from all effective corners; forming a transmission coefficient map based on the building material information of the indoor environment, thresholding the transmission coefficient map to locate areas where strong transmission occurs, and forming a transmission boundary point set from the boundary points of all strong transmission areas.

[0053] S20. Prune the diffraction point set in the initial discontinuous physical prior to update the discontinuous physical prior.

[0054] After identifying discontinuous physical priors, this invention prunes these priors to filter out redundant and invalid prior information, ensuring that the guidance information input to the model is highly relevant and low in noise. Specifically, this invention prunes the diffraction point set in the initial discontinuous physical priors, including: based on the geometric information of the indoor environment, removing diffraction points that cannot form an effective incident shadow region, and removing diffraction points whose regions on either side are geometrically connected and do not constitute a signal abrupt boundary. More specifically:

[0055] This invention is based on the principle that the pruning process of the effective incident shadow area follows the fundamental principles of geometric diffraction theory. The diffraction effect of a diffraction point is only non-negligible if it can form an effective incident shadow area. Therefore, this invention first calculates the transition contour, i.e., the line-of-sight boundary, between the indoor radio access point (AP) and the non-line-of-sight area. Based on this transition contour, it determines whether each diffraction point can form an effective blockage relative to the AP. If the diffraction point and its obstacle surface cannot form an effective incident shadow area, i.e., it is located within the entire line-of-sight area of ​​the AP, then the diffraction point is considered invalid and discarded.

[0056] To further refine the discontinuous physical priors, this invention uses the geometric information of the indoor environment to analyze the consistency of path loss on both sides of a diffraction point to eliminate redundant diffraction points. If the areas on both sides of a corner in the indoor environment experience similar penetration loss relative to the indoor radio access point (AP), meaning the path loss change of the signal propagation to both sides of the corner is smooth, then this diffraction point will not constitute a signal abrupt boundary. Such diffraction points are judged as redundant and removed from the diffraction point set.

[0057] S30. Generate a discontinuous physical prior map based on the indoor radio access point and the updated discontinuous physical prior.

[0058] The process of generating a discontinuous physical prior map in this embodiment of the invention includes: taking the indoor radio access point as the starting point, and using the direction from the indoor radio access point to each diffraction point and transmission boundary point in the updated discontinuous physical prior as the extension direction to form several rays; removing the portion between the indoor radio access point and each diffraction point and transmission boundary point in the updated discontinuous physical prior in each ray, and generating a discontinuous physical prior map based on the remaining portion of each ray. Figure 2 As shown, blue dots represent transmission boundary points, green dots represent diffraction points, and rays include rays passing through the indoor radio access point and each diffraction point as shown by the green lines, and rays passing through the indoor radio access point and each transmission boundary point as shown by the red lines. The removed portion of each ray is shown by the dashed lines, which means that the line segments between the indoor radio access point and each diffraction point and transmission boundary point in the updated discontinuous physical priors have been removed from each ray.

[0059] S40. Construct multimodal physical priors; wherein, multimodal physical priors include the coordinates of the indoor radio access point, the reflection coefficient map and transmission coefficient map characterizing the building materials at each location, and the discontinuous physical prior map.

[0060] In the physical condition input construction of the model, the embodiments of the present invention use physical information under four modes as prior information. It considers the coordinate position of the indoor radio access point, the reflection coefficient map and transmission coefficient map of the building materials at each location, and more importantly, the discontinuous physical prior map. The most critical and error-prone signal abrupt part is used as the prior condition, so that the model can perceive where the signal abrupt may occur during the generation process, thereby avoiding the oversmoothing problem caused by existing methods and significantly improving the modeling accuracy of physical details.

[0061] S50. After processing the multimodal physical priors through a multimodal fusion encoder, the processing result is used as the model input condition to input into the trained decoupled diffusion model to generate an indoor radio map.

[0062] This invention proposes a network model adapted for indoor radio map generation, as follows: Figure 3 As shown, it includes a multimodal fusion encoder and a decoupled diffusion model. The multimodal fusion encoder is introduced because the decoupled diffusion model expects to receive a three-channel image as a conditional input, while the multimodal physical prior constitutes a four-channel image. This invention designs a multimodal fusion encoder compatible with the decoupled diffusion model to convert the multimodal physical prior into a data form that matches the input of the decoupled diffusion model.

[0063] In this embodiment of the invention, the multimodal fusion encoder is a convolutional layer with a 1×1 kernel. Figure 3 The leftmost Conv utilizes a lightweight 1x1 convolution operation to perform a learnable fully connected linear transformation on each input channel at each pixel location. It learns to optimally combine, weight, and fuse four heterogeneous information types—reflection, transmission, AP proximity, and discontinuity prior—into three more information-dense and richer channel features.

[0064] In this embodiment of the invention, the decoupled diffusion model includes a conditional coding module, a noise feature extraction module, and a dual-target recovery and fusion module; wherein, the conditional coding module is used to extract noise features through an independent conditional encoder (…). Figure 3 The ConditionEncoder encodes the model input conditions (Cond) obtained from the multimodal fusion encoder to obtain conditional features; the noise feature extraction module is used to extract noise features from a U-Net encoder. Figure 3 The module consists of two parallel U-Net decoders (UNet Decoder1 and UNet Decoder2) that perform multi-scale feature extraction on the noisy indoor radio map (Input) of the current iteration to obtain noise features. The dual-target recovery and fusion module includes two parallel U-Net decoders (UNet Decoder1 and UNet Decoder2) that fuse noise features and conditional features based on the two parallel U-Net decoders through a cross-attention mechanism. Based on the fused features, the two parallel U-Net decoders recover two independent target outputs. One U-Net decoder recovers the prediction of the current noise component, and the other U-Net decoder recovers the direct prediction of the noise-free indoor radio map. Based on the two target outputs recovered by the dual-target recovery and fusion module, the noisy indoor radio map required for the next iteration is calculated and generated through a backdiffusion sampling algorithm until a clear indoor radio map is finally recovered.

[0065] The input conditions of the decoupled diffusion model are defined as a set. , Indicates the coordinates of the indoor radio access point. A diagram showing the reflectance coefficients of building materials at various locations. A graph showing the transmission coefficient of building materials at various locations. This represents a discontinuous physical prior graph, generated from discontinuous physical priors. The goal of this invention is to utilize network parameters as... Decoupling diffusion model It can learn from input conditions To estimate indoor radio map The mapping relationship, that is This decoupled diffusion model In the training dataset The training is performed on the above dataset, where N represents the total number of samples in the training dataset. The indoor radio map datasets used as training datasets include: an indoor radio map dataset generalized to antenna locations and an indoor radio map dataset generalized to zero-sample layout. Therefore, the optimization objective of this invention can be defined as finding a set of optimal network parameters. The goal is to minimize the mean square error between the indoor radio map predicted by the network and the actual indoor radio map. This optimization problem is formulated as follows:

[0066] ;

[0067] in, Indicates the first step in the training process The model input conditions corresponding to each training sample. Indicates the first step in the training process The objective function generates an indoor radio map corresponding to each training sample. By minimizing the prediction error, the decoupled diffusion model learns the complex electromagnetic wave propagation patterns indoors, thus generating an indoor radio map that is highly consistent with physical laws, given only the location of the access point (AP) and the building material properties.

[0068] Here, we assume a two-dimensional indoor area. A high-fidelity indoor radio map is constructed, the area being discretized into a uniform grid. The indoor radio map is represented as a Received Signal Strength Indicator (RSSI) matrix. .

[0069] The decoupling diffusion model in this invention is as follows: Figure 4 As shown, a high-fidelity radio map with multimodal physical information as the model condition input is recovered from a noisy image through an iterative denoising process. Figure 4 From left to right, The image contains random Gaussian noise. , For noisy indoor radio maps during the iterative denoising process. To reconstruct the generated high-fidelity radio map, the model dynamically focuses on the most important physical information of the current spatial location at each step of denoising. For example, when processing pixels near a corner, attention is focused on the input diffraction prior map, thus generating an indoor radio map highly consistent with physical laws. This design explicitly embeds the physical constraints of electromagnetic propagation into the learning process of the generative model, improving the accuracy of modeling physical details.

[0070] To verify the effectiveness of the indoor radio map construction method based on the physical enhancement diffusion model provided in this embodiment of the invention, the following experiments were conducted.

[0071] This invention comprehensively evaluates the proposed method using a publicly available indoor radio map dataset. This dataset, generated using ray tracing, includes 25 different indoor layouts and provides physically realistic indoor radio maps at the 3.5 GHz band, ignoring differences in antenna radiation patterns to isolate the effects of environmental propagation. It considers multiple reflections, transmissions, and diffractions. Each environment is discretized into a 0.25-meter grid, and the simulation considers up to 8 reflections, 10 transmissions, and 2 diffractions.

[0072] To systematically evaluate the generalization ability of the model, the evaluation protocol is divided into two scenarios:

[0073] Antenna location generalization (ALG): The training dataset contains all 25 scene layouts, but only 45 of the 50 AP locations are used for each layout; the test dataset uses the remaining 5 unseen AP locations in the same layout to evaluate the model’s adaptability to new deployments.

[0074] Zero-shot layout generalization (ZLG): The training dataset contains 20 scene layouts, each with 50 AP positions, while the test dataset contains 5 completely new scenes that the model has never seen during training, to evaluate the robustness of the model in unknown environments.

[0075] In the quantitative evaluation, multi-dimensional metrics were employed, including root mean square error (RMSE) and peak signal-to-noise ratio (PSNR), which measure pixel-level errors. Furthermore, to better capture the crucial perceptual quality and detail distribution in indoor radio maps, learned perceptual patch similarity (LPIPS) and Fréchet initiation distance (FID) were introduced as additional evaluation metrics. LPIPS measures the perceptual similarity between the generated map and the ground truth map in the depth feature space, while FID assesses the consistency of their overall distribution, which is essential for modeling fine-grained signal variations caused by multipath effects.

[0076] All experiments were run on an NVIDIA GeForce RTX 4090 operating system on Ubuntu. To comprehensively evaluate the effectiveness of the proposed method, denoted as iRadioDiff, this invention compares it with three representative deep learning architectures used for radio map construction: RadioUNet (CNN architecture), RME-GAN (Generative Adversarial Network), and SIP2Net (an enhanced U-Net architecture based on asymmetric convolution and ASPP). The metrics comparison for each architecture is shown in Table 1.

[0077] Table 1 Comparison of metrics under different architectures

[0078]

[0079] As shown in Table 1, in the antenna position generalization (ALG) scenario, the RMSE of this invention is as low as 6.357, a 32% reduction compared to the second-best RadioUNet (9.349), representing a significant leap in accuracy. Simultaneously, its PSNR reaches a high of 32.24, and it significantly outperforms other methods in the perception metrics LPIPS (0.2742) and FID (145.2), indicating that the map generated by this invention not only has lower pixel-level errors but also a more realistic overall structure and detail distribution. In the more challenging zero-shot layout generalization (ZLG) scenario, this invention maintains strong performance. Despite facing a completely new environment, its RMSE remains at 7.010, and its PSNR at 31.45, both superior to all compared methods.

[0080] Experimental results show that the iRadioDiff method proposed in this invention achieves near-optimal performance across all evaluation metrics in both generalization scenarios, comprehensively surpassing existing deep learning methods.

[0081] Furthermore, the effectiveness of the iRadioDiff method proposed in this invention in localization tasks was verified. Localization verification was performed using the KNN algorithm with K=5, and the results are shown in Table 2.

[0082] Table 2. Location Task Results

[0083]

[0084] As can be seen from Table 2, in the positioning task, the iRadioDiff proposed in this invention is the only method that can stably achieve sub-10-meter positioning accuracy. Its positioning errors in ALG and ZLG scenarios reached 7.860 meters and 8.530 meters, respectively. This shows that the invention can accurately infer the signal distribution of new AP locations in known scenarios. The physical laws learned by this invention have strong transferability and can be effectively applied to new building environments. It has outstanding generalization ability, and the generated radio maps have excellent value in practical applications.

[0085] in addition, Figure 5 and Figure 6 The visualization results of each method are shown in the antenna position generalization scenario and the zero-shot layout generalization scenario, respectively. Figure 5 and Figure 6 The visualization results further highlight the advantages of the iRadioDiff method proposed in this invention: it can more effectively capture fine structural details and signal discontinuities caused by multipath propagation, diffraction, and transmission, thereby achieving clearer radio map reconstruction with richer physical details. Especially in areas where signal strength changes drastically, such as corners and doorways, this invention can accurately model these fine-grained changes, while traditional methods such as RadioUNet exhibit obvious oversmoothing and fail to capture these key physical information.

[0086] To verify the effectiveness of the "physical information prior" proposed in this invention, an ablation study was conducted. In this ablation study, the method using the complete physical information prior of this invention is labeled "w / Physics". The ablation experimental version that removed all physical-related prior inputs (i.e., discontinuous physical prior maps) and made the model only receive the reflection coefficient map, transmission coefficient map, and AP location is labeled "w / o Physics". The ablation study results are shown in Table 3.

[0087] Table 3 Ablation Study of Physical Information Priors

[0088]

[0089] As shown in Table 3, removing the physical priors (“w / o Physics”) significantly reduced all performance metrics of the model, including indoor radio map construction accuracy and positioning accuracy. For example, in the ALG task, the RMSE deteriorated from 6.357 to 9.619, and the positioning error also increased accordingly. This result strongly demonstrates that the strategy proposed in this invention, which incorporates physical priors such as material electromagnetic properties, diffraction, and transmission boundaries into the diffusion model, is key to achieving high fidelity and strong generalization ability.

[0090] The above experimental results fully verify that the physical enhancement diffusion model of the present invention not only greatly improves the reconstruction accuracy of indoor radio maps, but also maintains its excellence in complex environments such as unknown antenna locations and new indoor layouts, making it a powerful tool for high-precision positioning and network optimization in next-generation communication networks.

[0091] In summary, the indoor radio map construction method based on a physical enhancement diffusion model proposed in this invention overcomes the problems of high computational latency, reliance on field measurements, and insufficient modeling of complex indoor environments in existing indoor radio map construction technologies. It is a high-fidelity indoor radio map construction method that requires no field measurements, can generate maps quickly, and has the following significant advantages: By using material electromagnetic properties and key physical structures (diffraction points, transmission boundary points) as priors, the indoor radio map generated by this invention can highly realistically reproduce the details of signal propagation in complex indoor environments, especially in boundary areas where signals change drastically, with accuracy far exceeding existing technologies; the method proposed in this invention requires no field measurements. This invention generates high-precision indoor radio maps based solely on building transmission coefficient maps, reflection coefficient maps, individual access point (AP) locations, and discontinuous physical prior maps, significantly reducing deployment costs and time and supporting rapid response to dynamic environmental changes. The invention employs a decoupled diffusion model, which learns universal electromagnetic propagation physics rather than specific scenario-specific data distributions. Therefore, it effectively generalizes to entirely new indoor layouts and AP locations, exhibiting strong robustness. Experiments demonstrate that using the indoor radio map generated by this invention as a fingerprint database achieves sub-10-meter indoor positioning accuracy, providing a solid technical foundation for various downstream applications requiring precise location information.

[0092] Secondly, please see Figure 7 This invention provides an indoor radio map construction device based on a physical augmentation diffusion model, the indoor radio map construction device comprising:

[0093] The first construction module is used to construct the initial discontinuous physical prior based on the geometric information of the indoor environment and the building material information; the initial discontinuous physical prior includes the diffraction point set and the transmission boundary point set;

[0094] The pruning module is used to prune the diffraction point set in the initial discontinuous physical prior in order to update the discontinuous physical prior.

[0095] The first generation module is used to generate a discontinuous physical prior map based on the indoor radio access point and the updated discontinuous physical prior.

[0096] The second building module is used to construct multimodal physical priors; among which, multimodal physical priors include the coordinate location of the indoor radio access point, the reflection coefficient map and transmission coefficient map characterizing the building materials at each location, and the discontinuous physical prior map;

[0097] The second generation module is used to process the multimodal physical priors through a multimodal fusion encoder, and then use the processing results as input conditions to the trained decoupled diffusion model to generate an indoor radio map.

[0098] As the apparatus embodiment of the second aspect is basically similar to the method embodiment of the first aspect, the description is relatively simple, and relevant details can be found in the description of the method embodiment of the first aspect.

[0099] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0100] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the specification and accompanying drawings, will understand and implement other variations of the disclosed embodiments in carrying out the claimed invention. In the specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. While certain measures are described in different embodiments, this does not mean that these measures cannot be combined to produce good results.

[0101] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for constructing indoor radio maps based on a physical augmentation diffusion model, characterized in that, The indoor radio map construction method includes: Based on the geometric information of the indoor environment and the building material information, an initial discontinuity physical prior is constructed. The initial discontinuity physical prior includes a diffraction point set and a transmission boundary point set. Specifically, constructing the initial discontinuity physical prior based on the geometric information of the indoor environment and the building material information includes: using a geometric neighborhood analysis algorithm to locate the effective corners where diffraction occurs based on the geometric information of the indoor environment, and forming a diffraction point set from all effective corners; and generating a transmission coefficient map based on the building material information of the indoor environment, performing thresholding on the transmission coefficient map to locate the areas where strong transmission occurs, and forming a transmission boundary point set from the boundary points of all strong transmission areas. The diffraction point set in the initial discontinuous physical prior is pruned to update the discontinuous physical prior; wherein, the pruning of the diffraction point set in the initial discontinuous physical prior includes: removing diffraction points in the diffraction point set that cannot form an effective incident shadow area based on the geometric information of the indoor environment, and removing diffraction points in the diffraction point set whose regions on both sides are geometrically connected and do not constitute a signal abrupt boundary. Based on the indoor radio access point and the updated discontinuous physical prior, a discontinuous physical prior map is generated. The process of generating the discontinuous physical prior map includes: taking the indoor radio access point as the starting point and using the directions from the indoor radio access point to each diffraction point and transmission boundary point in the updated discontinuous physical prior as the extension directions to form several rays; removing the portion between the indoor radio access point and each diffraction point and transmission boundary point in the updated discontinuous physical prior in each ray, and generating a discontinuous physical prior map based on the remaining portion of each ray. Construct a multimodal physical prior; the multimodal physical prior includes the coordinate location of the indoor radio access point, the reflection coefficient map and transmission coefficient map characterizing the building materials at each location, and the discontinuous physical prior map; The multimodal physical priors are processed by a multimodal fusion encoder, and the processing result is used as the model input condition to input into the trained decoupled diffusion model to generate an indoor radio map.

2. The indoor radio map construction method based on the physical augmentation diffusion model according to claim 1, characterized in that, The multimodal fusion encoder is a convolutional layer with a 1×1 kernel.

3. The method for constructing indoor radio maps based on a physical augmentation diffusion model according to claim 1, characterized in that, The decoupled diffusion model includes a conditional coding module, a noise feature extraction module, and a dual-target recovery and fusion module; among which, The conditional encoding module is used to encode the model input conditions obtained by the multimodal fusion encoder through an independent conditional encoder to obtain conditional features; The noise feature extraction module is used to extract noise features from the noisy indoor radio map of the current iteration using a U-Net encoder at multiple scales. The dual-target recovery and fusion module includes two parallel U-Net decoders. Based on the two parallel U-Net decoders, noise features and conditional features are fused through a cross-attention mechanism. According to the fused features, the two parallel U-Net decoders recover two independent target outputs. One U-Net decoder recovers the prediction of the current noise component, and the other U-Net decoder recovers the direct prediction of the noise-free indoor radio map. Based on the two target outputs recovered by the dual-target recovery and fusion module, the noisy indoor radio map required for the next iteration is calculated and generated through a back-diffusion sampling algorithm until a clear indoor radio map is finally recovered.

4. The method for constructing indoor radio maps based on a physical augmentation diffusion model according to claim 1, characterized in that, The indoor radio map datasets used in the training of the decoupled diffusion model include an indoor radio map dataset generalized to antenna location and an indoor radio map dataset generalized to zero-sample layout.

5. An indoor radio map construction device based on a physical augmentation diffusion model, characterized in that, The indoor radio map building device includes: The first construction module is used to construct an initial discontinuity physical prior based on the geometric information of the indoor environment and the building material information. The initial discontinuity physical prior includes a diffraction point set and a transmission boundary point set. Specifically, constructing the initial discontinuity physical prior based on the geometric information of the indoor environment and the building material information includes: using a geometric neighborhood analysis algorithm to locate the effective corners where diffraction occurs based on the geometric information of the indoor environment, and forming a diffraction point set from all effective corners; forming a transmission coefficient map based on the building material information of the indoor environment, thresholding the transmission coefficient map to locate the areas where strong transmission occurs, and forming a transmission boundary point set from the boundary points of all strong transmission areas. The pruning module is used to prune the diffraction point set in the initial discontinuous physical prior to update the discontinuous physical prior; wherein, the pruning of the diffraction point set in the initial discontinuous physical prior includes: removing diffraction points in the diffraction point set that cannot form an effective incident shadow area based on the geometric information of the indoor environment, and removing diffraction points in the diffraction point set whose two sides are geometrically connected and do not constitute a signal abrupt boundary; The first generation module is used to generate a discontinuous physical prior map based on the indoor radio access point and the updated discontinuous physical prior. The process of generating the discontinuous physical prior map includes: taking the indoor radio access point as the starting point and taking the direction from the indoor radio access point to each diffraction point and transmission boundary point in the updated discontinuous physical prior as the extension direction to form several rays; removing the part between the indoor radio access point and each diffraction point and transmission boundary point in the updated discontinuous physical prior in each ray, and generating a discontinuous physical prior map based on the remaining part of each ray. The second building module is used to construct multimodal physical priors; among which, multimodal physical priors include the coordinate location of the indoor radio access point, the reflection coefficient map and transmission coefficient map characterizing the building materials at each location, and the discontinuous physical prior map; The second generation module is used to process the multimodal physical priors through a multimodal fusion encoder, and then use the processing results as input conditions to the trained decoupled diffusion model to generate an indoor radio map.

Images