Radio map construction method based on physical manifold alignment for few-shot diffusion model

Abstract

This invention discloses a method for constructing radio maps based on a few-shot diffusion model using physical manifold alignment, belonging to the field of radio map construction technology. The method includes: inputting a Gaussian random noise map and corresponding first environmental conditions into a trained radio map generation network; wherein the first environmental conditions corresponding to the Gaussian random noise map include a first environmental occupancy matrix and first transmitter location information; the radio map generation network is constructed according to the U-Net architecture and trained using a main-path radio map dataset, a multi-path radio map dataset, and a preset loss function; the preset loss function includes directional consistency loss; and generating a corresponding high-fidelity radio map through the reverse generation process of the trained radio map generation network. This invention achieves high-precision modeling of complex wireless propagation environments under few-shot conditions, exhibiting significant technological advancement, engineering practicality, and broad application prospects.

Description

Technical Field

[0001] This invention belongs to the field of radio map construction technology, specifically relating to a method for constructing radio maps using a few-sample diffusion model based on physical manifold alignment. Background Technology

[0002] As a core infrastructure for 6G network intelligence, radio maps support low-latency applications (such as UAV relay planning and digital twins) by decoupling channel state acquisition from real-time pilot overhead. However, existing technologies face a dual bottleneck: First, traditional electromagnetic solvers (such as high-precision ray tracing) require minutes of latency to generate a single-scene radio map (generating 50,000 samples would take more than 800 hours), which cannot meet real-time requirements. Second, the number of labeled samples required for data-driven model training is too large, with data acquisition costs exceeding $100 per hour. Furthermore, simulation data, due to the simplification of propagation physics (such as ignoring higher-order reflection / diffraction effects), may result in distribution mismatch and lack of physical consistency when reconstructing the real environment, making it impossible to accurately capture non-stationary diffraction edge effects.

[0003] The mainstream technical approaches for constructing high-fidelity radio maps currently fall into the following three categories: The first category is traditional electromagnetic solution techniques, including full-wave methods and ray tracing, which generate physically consistent radio maps by rigorously solving electromagnetic propagation equations or simulating ray propagation paths. Among these, the principal path model is widely used for generating large-scale datasets due to its computational efficiency, but it ignores multipath effects. Intelligent ray tracing can capture multipath effects, but it has extremely high computational overhead.

[0004] The second category is data-driven generative models, including convolutional neural networks, generative adversarial networks, and diffusion models. These models quickly generate radio maps by learning the distribution characteristics of training data. Among them, diffusion models perform well in terms of structural fidelity thanks to their iterative denoising mechanism, but they still rely on large-scale labeled data.

[0005] The third category is the physically informed learning model: by embedding electromagnetic physical constraints (such as Helmholtz equations or boundary conditions) into the model training, the physical consistency of the generated results is improved. Among them, although this type of model improves generalization, it still requires a large amount of high-quality physically consistent data to support it, and it does not solve the cross-domain adaptation problem in scenarios with few samples.

[0006] Based on the above discussion, existing technical solutions to the problem of radio map construction all have some significant limitations and challenges.

[0007] Traditional electromagnetic solving techniques have extremely low computational efficiency, failing to meet the demands of real-time deployment or large-scale scenarios. While high-precision electromagnetic solving techniques can generate multipath radio maps, the high cost of sample generation makes it difficult to support the training of data-driven models. Meanwhile, data-driven generation models heavily rely on massive amounts of labeled data. When faced with multipath effects in real-world environments, the simplification of training data leads to distribution mismatch, resulting in a lack of physical consistency in the generated results. Advanced architectures such as diffusion models have even higher data requirements and are prone to issues like pattern illusion and blurred edges in scenarios with few samples. Although physically informed learning models incorporate physical constraints, they still have not overcome the data dependency bottleneck, making it difficult to achieve efficient transfer from main path to multipath radio maps under conditions with few samples. Summary of the Invention

[0008] To address the aforementioned problems in the prior art, this invention provides a method for constructing radio maps using a few-sample diffusion model based on physical manifold alignment.

[0009] The technical problem to be solved by this invention is achieved through the following technical solution: In a first aspect, the present invention provides a method for constructing radio maps using a few-sample diffusion model based on physical manifold alignment, the method comprising: A Gaussian random noise map and its corresponding first environmental condition are input into a trained radio map generation network. The first environmental condition, corresponding to the Gaussian random noise map, includes a first environmental occupancy matrix and first transmitter location information. The radio map generation network is constructed based on the U-Net architecture. The network is pre-trained using a pre-built main-path radio map dataset and then further trained using a pre-built multi-path radio map dataset and a preset loss function. The preset loss function includes direction consistency loss. The high-fidelity radio map is generated based on the Gaussian random noise map through the reverse generation process of the trained radio map generation network.

[0010] Optionally, the process of constructing the main path radio map dataset and the multipath radio map dataset includes: The discretized two-dimensional planar region is divided into multiple grid regions of the same size; each grid region has the same spatial resolution and the path loss is constant within each grid region. Set the second transmitter position information of the transmitter in the discretized two-dimensional plane region, and set the second environmental geometry according to the binary occupancy matrix; The second transmitter location information and the second environmental geometry are used as the second environmental conditions; Based on the multiple grid regions, the second environmental conditions, and the radio propagation mechanism, the main path radio map dataset and the multipath radio map dataset are obtained.

[0011] Optionally, the forward diffusion process of the radio map generation network is represented as follows: ; in, Indicates the first Gaussian random noise samples at each time step Indicates the first Control signal attenuation term for each time step Indicates the first Control noise injection intensity at each time step Indicates the first A standard Wiener process with one time step.

[0012] Optionally, the reverse generation process of the radio map generation network is represented as follows: ; in, Represents Gaussian random noise samples gradient operator, Represents Gaussian random noise samples The probability density function.

[0013] Optionally, the preset loss function is expressed as follows: ; in, To rebuild the losses, For loss of directional consistency, The weighting coefficients for the direction consistency loss are... Trainable parameters for the radio map generation network.

[0014] Optionally, the training process of the radio map generation network includes: The radio map generation network is pre-trained using the pre-built main path radio map dataset to obtain the pre-trained radio map generation network. After freezing the encoder weights of the pre-trained radio map generation network, a second training is performed based on the pre-constructed multipath radio map dataset and a preset loss function to obtain the trained radio map generation network.

[0015] Secondly, the present invention provides a radio map construction apparatus based on a few-sample diffusion model with physical manifold alignment, the apparatus comprising: An input module is used to input a Gaussian random noise map and corresponding environmental conditions into a trained radio map generation network; wherein, the environmental conditions corresponding to the Gaussian random noise map include the corresponding environmental occupancy matrix and transmitter location information; the radio map generation network is constructed according to the U-Net architecture; the radio map generation network is pre-trained based on a pre-built main-path radio map dataset and then secondary-trained based on a pre-built multi-path radio map dataset and a preset loss function; the preset loss function includes direction consistency loss; The generation module is used to generate a corresponding high-fidelity radio map based on the Gaussian random noise map through the reverse generation process of the trained radio map generation network.

[0016] Thirdly, the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the steps of the method for constructing a few-sample diffusion model radio map based on physical manifold alignment as described in the first aspect above.

[0017] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having computer program instructions stored thereon, which, when executed by a processor, implement the steps of the method for constructing a few-sample diffusion model radio map based on physical manifold alignment as described in the first aspect of the present invention.

[0018] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: In the above technical solution, the present invention obtains a radio map generation network through two-stage training based on the main path radio map dataset, the multipath radio map dataset, and the direction consistency loss. This network can generate high-fidelity radio maps while ensuring the physical consistency and spatial structural integrity of the generated radio maps. It achieves high-precision modeling of complex wireless propagation environments under conditions of few samples, and has significant technical advantages, engineering practicality, and broad application prospects.

[0019] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0020] Figure 1 This is a flowchart of a method for constructing a few-sample diffusion model radio map based on physical manifold alignment, provided by an embodiment of the present invention; Figure 2This is a comparison diagram of the generation effects of different radio map construction methods in static scenes provided by embodiments of the present invention; Figure 3 This is a comparison diagram of the generation effects of different radio map construction methods in dynamic scenes provided by embodiments of the present invention; Figure 4 This is a block diagram of a few-sample diffusion model radio map construction device based on physical manifold alignment provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0021] 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.

[0022] Figure 1 This is a flowchart of a few-shot diffusion model radio map construction method based on physical manifold alignment provided by an embodiment of the present invention, such as... Figure 1 As shown, the method may include the following steps: S101. Input the Gaussian random noise map and the corresponding first environmental conditions into the trained radio map generation network; wherein, the first environmental conditions corresponding to the Gaussian random noise map include the first environmental occupancy matrix and the first transmitter position information; the radio map generation network is constructed according to the U-Net architecture; the radio map generation network is pre-trained based on the pre-built main path radio map dataset and then secondary-trained based on the pre-built multipath radio map dataset and the preset loss function; the preset loss function includes direction consistency loss.

[0023] Optionally, the construction process of the main path radio map dataset and the multipath radio map dataset includes: The discretized two-dimensional planar region is divided into multiple grid regions of the same size; each grid region has the same spatial resolution and the path loss is constant within each grid region. Set the second transmitter position information in the discretized two-dimensional plane region, and set the second environmental geometry according to the binary occupancy matrix; The second transmitter's location information and the second environmental geometry are used as the second environmental conditions; Based on multiple grid regions, secondary environmental conditions, and radio propagation mechanisms, the main path radio map dataset and the multipath radio map dataset are obtained.

[0024] Specifically, a discretized two-dimensional planar region is divided into parts of size . A regular grid region. Each grid region corresponds to a spatial resolution of [missing information]. The area is defined, and it is assumed that the path loss is approximately constant within this grid area. The radio map is represented as a matrix. ,in, Representing the real number field, at carrier frequency Next Line number The path loss value at the column grid position is expressed as The location information of the second transmitter is as follows: The second environmental geometry is composed of a binary occupancy matrix. The description assumes that an obstacle completely blocks the direct path of propagation. For example, based on different radio propagation mechanisms, radio maps are divided into two categories: main path radio maps (MP-RM / SRM), considering only the strongest propagation path, for each group... The received power is calculated using the primary path model, generating the corresponding primary path radio map. This process is repeated to form a primary path radio map dataset containing multiple data pairs. Correspondingly, a high-precision ray tracing model is used to generate the corresponding multipath radio map, referencing the above process. Specifically, the primary path radio map (SRM / MP-RM) is generated based on the Dominant Path Model (DPM). This model calculates only the strongest propagation path between the transmitter and receiver, ignoring all reflection and diffraction effects, resulting in extremely high computational efficiency. It is used to construct a primary path radio map dataset containing a large number of samples. The multipath radio map (MRM / MU-RM) is generated based on the high-precision Intelligent Ray Tracing (IRT) algorithm. This algorithm simulates the multipath propagation of electromagnetic waves in complex urban environments, including reflection (supporting up to 2 times) and diffraction (supporting up to 1 time). The results are used as a baseline to construct a multipath radio map dataset with fewer samples. The original dataset does not contain noise.

[0025] Alternatively, since the forward diffusion process defines how to obtain a clear radio map... Gradually transform into Gaussian random noise samples This allows U-Net to learn the real electromagnetic map features under different noise levels, providing a reference frame for the network's learning. The forward diffusion process of the radio map generation network is represented by the following stochastic differential equation: ; in, Indicates the first Gaussian random noise samples at each time step Indicates the first Control signal attenuation term for each time step Indicates the first Control noise injection intensity at each time step Indicates the first A standard Wiener process with one time step.

[0026] Optionally, the reverse generation process defines how to progressively recover a radio map from Gaussian random noise samples. By solving the inverse time equation, U-Net can reconstruct an electromagnetic map conforming to electromagnetic propagation characteristics from the noise according to the forward diffusion process. This invention models the radio map generation process, and the reverse generation process is implemented by solving the corresponding inverse-time stochastic differential equation, as shown below: ; in, express The gradient operator is parameterized and learned by the neural network. Indicates the first time step The probability density function.

[0027] Specifically, the forward diffusion process defines how to progressively transform a clear radio map into Gaussian random noise samples. This equation can calculate the noisy map at any time step, allowing U-Net to learn the true characteristics of radio maps at different noise levels, providing a reference frame for the network's learning. The backward generation process defines how to progressively recover the radio map from pure noise. By solving this inverse time, the model can use what it has learned to reconstruct a radio map from the noise that conforms to the electromagnetic propagation characteristics.

[0028] It is understood that the present invention uses a conditional diffusion generation model as the radio map generation network. The radio map generation network is built on the U-Net architecture. The U-Net network consists of an encoder, a decoder and multi-layer skip connections. The encoder gradually extracts multi-scale spatial features, and the decoder gradually recovers the spatial structure of the radio map under the constraints of the second environmental condition. The time step embedding is injected into multiple layers of the radio map generation network through sinusoidal position encoding to achieve modeling of different diffusion stages.

[0029] Optionally, the training process for the radio map generation network includes: The radio map generation network was pre-trained using a pre-built main path radio map dataset to obtain the pre-trained radio map generation network. After freezing the encoder weights of the pre-trained radio map generation network, a second training is performed based on the pre-built multipath radio map dataset and the preset loss function to obtain the trained radio map generation network.

[0030] Understandably, the input to this radio map generation network during training includes Gaussian random noise samples from the main path radio map dataset and the multipath radio map dataset. Time step This is used to represent the noise scale during the diffusion process; second environmental condition. Second environmental conditions Including the second environmental geometry and the location information of the second transmitter .

[0031] Understandably, the pre-training phase involves training the radio map generation network using Gaussian random noise samples from a large number of main-path radio map datasets to learn a stable spatial propagation skeleton. The few-sample fine-tuning phase involves fine-tuning the network using a small number of Gaussian random noise samples from a few multipath radio map datasets, while freezing some network parameters, and introducing directional consistency loss for regularization. Specifically, the encoder weights that extract environmental geometry and large-scale attenuation features from the pre-trained radio map generation network are frozen. Fine-tuning is then performed on specific layers of the decoder that model multipath reflection and diffraction residuals, as well as residual connection modules, primarily focusing on the decoder part of the pre-trained radio map generation network.

[0032] Understandably, according to electromagnetic propagation theory, multipath radio maps can be decomposed into principal path components determined by environmental geometry and nonlinear multipath residuals caused by complex environments. In the feature space, the migration from the principal path domain to the multipath domain corresponds to a biased manifold displacement. Due to the extremely small amount of data during fine-tuning with few samples, the model is prone to "rotational uncertainty" in the update direction, leading to physically incongruous interference noise in the generated images.

[0033] To ensure the physical consistency of the generated results during the few-sample adaptation process, this invention introduces a directional consistency loss. This loss constrains the generated features to change only along the principal direction related to multipath injection during the migration from the principal path domain to the multipath domain, suppressing irrelevant perturbations orthogonal to this direction. The directional consistency loss is viewed as a lightweight manifold projection operator used to eliminate rotational uncertainties in the feature space and ensure the uniqueness of the cross-domain displacement direction. The directional consistency loss is constructed by measuring the consistency between the gradient of the score function predicted by the network and the reference direction of the physical manifold. The specific expression is:

[0034] in, This represents the feature update vector predicted by the radio map generation network at the current time step. This represents a predefined physical manifold reference vector, which is obtained by calculating the average displacement gradient in the feature space between a small number of high-fidelity multipath samples and their corresponding principal path samples. It represents the main physical direction of the evolution from principal path propagation to multipath propagation. Represents the dimension index of the vector.

[0035] Orientation consistency loss forces the model to update parameters only along physically relevant directions during fine-tuning, suppressing invalid random perturbations in orthogonal directions. With orientation consistency loss, the model can accurately capture the diffraction texture of building edges even with only 10-50 samples, without disrupting the global energy distribution skeleton learned during pre-training. Experiments validated this training strategy's advantage of maintaining structural consistency and generation stability even with very few samples. Finally, the pre-defined loss function is expressed as follows: ; in, To rebuild the losses, For loss of directional consistency, The weighting coefficients for the direction consistency loss are... The trainable parameters for the radio map generation network are determined by adjusting these parameters to minimize the weighted sum of reconstruction loss and orientation consistency loss, thereby obtaining the optimal model parameters.

[0036] S102. Through the reverse generation process of the trained radio map generation network, a corresponding high-fidelity radio map is generated based on the Gaussian random noise map.

[0037] In one implementation, Figure 2 This is a comparison chart showing the generation effects of different radio map construction methods in static scenes, as provided in an embodiment of the present invention. Figure 3 This is a comparison diagram of the generation effects of different radio map construction methods in dynamic scenes provided by embodiments of the present invention, such as... Figure 2 and Figure 3As shown, the diagram illustrates the comparison of generation effects of different radio map construction methods in static and dynamic scenes under the same environmental conditions. Specifically, these include RadioUNet (a fast radio map estimation model using convolutional neural networks), PhyRMDM (a physically aligned radio map diffusion model), RadioDiff (a generative electromagnetic map construction method based on a denoising diffusion model), RME-GAN (a radio map estimation framework based on conditional generative adversarial networks), and the present invention (RadioDiff-FS). Each row represents a specific scene, corresponding to different building layouts and transmitter locations; each column represents the effect of radio maps generated by different methods, and the last column is the ground truth. Compared to existing radio map construction methods, the method proposed in this invention can maintain more continuous and consistent spatial structure characteristics in complex multipath propagation areas of the radio propagation environment, thus more accurately reflecting the distribution patterns of radio propagation in complex environments. Especially in dynamic scenarios, when environmental geometry or propagation conditions change, the stability of the results generated by existing methods decreases significantly. However, the method of this invention, due to the introduction of physical constraints and cross-domain manifold alignment mechanisms in the modeling process, can still maintain the rationality and consistency of the overall structure of the radio map, demonstrating stronger environmental adaptability and generalization ability.

[0038] This invention utilizes a radio map generation network trained in two stages using a main path radio map dataset, a multipath radio map dataset, and a direction consistency loss. This network can generate high-fidelity radio maps while ensuring the physical consistency and spatial structural integrity of the generated radio maps. It achieves high-precision modeling of complex wireless propagation environments under limited sample conditions, demonstrating significant technological advancement, engineering practicality, and broad application prospects.

[0039] Figure 4 This is a block diagram of a few-sample diffusion model radio map construction device based on physical manifold alignment provided in an embodiment of the present invention, such as... Figure 4 As shown, the device 400 may include: Input module 401 is used to input a Gaussian random noise map and the corresponding environmental conditions into the trained radio map generation network; wherein, the environmental conditions corresponding to the Gaussian random noise map include the corresponding environmental occupancy matrix and transmitter position information; the radio map generation network is constructed according to the U-Net architecture; the radio map generation network is pre-trained based on a pre-built main path radio map dataset and then secondary-trained based on a pre-built multipath radio map dataset and a preset loss function; the preset loss function includes orientation consistency loss; The generation module 402 is used to generate a corresponding high-fidelity radio map based on the Gaussian random noise map by reverse generation process of the trained radio map generation network.

[0040] It is understood that the device embodiments are basically similar to the method embodiments, so the description is relatively simple, and relevant parts can be referred to in the description of the method embodiments.

[0041] Figure 5 This is a schematic diagram of an electronic device provided in an embodiment of the present invention, such as... Figure 5 As shown, it includes a processor 501, a communication interface 502, a memory 503, and a communication bus 504, wherein the processor 501, the communication interface 502, and the memory 503 communicate with each other through the communication bus 504. Memory 503 is used to store computer programs; When the processor 501 executes the program stored in the memory 503, it implements the steps of the above-described method for constructing radio maps using a few-sample diffusion model based on physical manifold alignment.

[0042] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus.

[0043] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0044] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0045] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0046] The present invention also provides a computer-readable storage medium. A computer program is stored in the computer-readable storage medium, and when executed by a processor, the computer program implements the steps of the above-described method for constructing a few-sample diffusion model radio map based on physical manifold alignment.

[0047] Optionally, the computer-readable storage medium may be non-volatile memory (NVM), such as at least one disk storage device.

[0048] Optionally, the computer-readable storage device may also be at least one storage device located remotely from the aforementioned processor.

[0049] In another embodiment of the present invention, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform the steps of the above-described method for constructing radio maps using a few-sample diffusion model based on physical manifold alignment.

[0050] It should be noted that the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention.

[0051] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0052] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings and the disclosure in carrying out the claimed invention. In the description of the invention, the word "comprising" does not exclude other components or steps, "a" or "an" does not exclude a plurality, and "a plurality" means two or more, unless otherwise explicitly specified. Furthermore, while different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce good results.

[0053] The method provided in this invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc. No limitation is made herein; any electronic device that can implement this invention falls within the protection scope of this invention.

[0054] For the embodiments of the device / electronic device / storage medium, since they are basically similar to the method embodiments, the description is relatively simple, and relevant parts can be referred to in the description of the method embodiments.

[0055] It should be noted that the device, electronic device, and storage medium in the embodiments of the present invention are respectively the device, electronic device, and storage medium for constructing radio maps using the few-shot diffusion model based on physical manifold alignment. Therefore, all embodiments of the few-shot diffusion model radio map construction method based on physical manifold alignment are applicable to the device, electronic device, and storage medium, and can achieve the same or similar beneficial effects.

[0056] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus (electronic devices), or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects, all of which are collectively referred to herein as "modules" or "systems." Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The computer program may be stored / distributed in a suitable medium, provided with or as part of other hardware, or may take other distribution forms, such as via the Internet or other wired or wireless telecommunications systems.

[0057] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (electronic devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0058] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0059] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0060] 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 physical manifold alignment based few-shot diffusion model radio map construction method, characterized in that, The method includes: A Gaussian random noise map and its corresponding first environmental condition are input into a trained radio map generation network. The first environmental condition, corresponding to the Gaussian random noise map, includes a first environmental occupancy matrix and first transmitter location information. The radio map generation network is constructed based on the U-Net architecture. The network is pre-trained using a pre-built main-path radio map dataset and then further trained using a pre-built multi-path radio map dataset and a preset loss function. The preset loss function includes direction consistency loss. The high-fidelity radio map is generated based on the Gaussian random noise map through the reverse generation process of the trained radio map generation network.

2. The physical manifold alignment based few-shot diffusion model radio map construction method according to claim 1, characterized in that, The construction process of the main path radio map dataset and the multipath radio map dataset includes: The discretized two-dimensional planar region is divided into multiple grid regions of the same size; each grid region has the same spatial resolution and the path loss is constant within each grid region. Set the second transmitter position information of the transmitter in the discretized two-dimensional plane region, and set the second environmental geometry according to the binary occupancy matrix; The second transmitter location information and the second environmental geometry are used as the second environmental conditions; Based on the multiple grid regions, the second environmental conditions, and the radio propagation mechanism, the main path radio map dataset and the multipath radio map dataset are obtained.

3. The physical manifold alignment based few-shot diffusion model radio map construction method according to claim 1, characterized in that, The forward diffusion process of the radio map generation network is represented as follows: ； in, Indicates the first Gaussian random noise samples at each time step Indicates the first Control signal attenuation term for each time step Indicates the first Control noise injection intensity at each time step Indicates the first A standard Wiener process with one time step.

4. The physical manifold alignment based few-shot diffusion model radio map construction method according to claim 3, characterized in that, The reverse generation process of the radio map generation network is represented as follows: ； in, Represents Gaussian random noise samples gradient operator, Represents Gaussian random noise samples The probability density function.

5. The method for constructing a few-shot diffusion model radio map based on physical manifold alignment according to claim 1, characterized in that, The preset loss function is expressed as follows: ； wherein, is a reconstruction loss, is a direction consistency loss, is a weight coefficient of the direction consistency loss, are trainable parameters of the radio map generating network.

6. The physical manifold alignment based few-shot diffusion model radio map construction method according to claim 1, characterized in that, The training process of the radio map generation network includes: The radio map generation network is pre-trained using the pre-built main path radio map dataset to obtain the pre-trained radio map generation network. After freezing the encoder weights of the pre-trained radio map generation network, a second training is performed based on the pre-constructed multipath radio map dataset and a preset loss function to obtain the trained radio map generation network.

7. A physical manifold alignment based few-shot diffusion model radio map construction apparatus, characterized in that, The device includes: An input module is used to input a Gaussian random noise map and corresponding environmental conditions into a trained radio map generation network; wherein, the environmental conditions corresponding to the Gaussian random noise map include the corresponding environmental occupancy matrix and transmitter location information; the radio map generation network is constructed according to the U-Net architecture; the radio map generation network is pre-trained based on a pre-built main-path radio map dataset and then secondary-trained based on a pre-built multi-path radio map dataset and a preset loss function; the preset loss function includes direction consistency loss; The generation module is used to generate a corresponding high-fidelity radio map based on the Gaussian random noise map through the reverse generation process of the trained radio map generation network.

8. An electronic device, comprising: include: processor; Memory used to store processor-executable instructions; The processor is configured to execute the executable instructions to implement the steps of the method for constructing a radio map based on a physical manifold alignment and a few-sample diffusion model according to any one of claims 1-6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, The program instructions, when executed by the processor, implement the steps of the method for constructing a radio map based on a physical manifold alignment and a few-sample diffusion model according to any one of claims 1-6.

Images