Fast wireless channel modeling method and system based on voxelized radiation field

Abstract

The application provides a kind of fast wireless channel modeling method and system based on voxelized radiation field, comprising: the space to be modeled is divided into three-dimensional voxel grid, each voxel stores body density and characteristic vector;Ray sampling is carried out along the direction of receiver Rx to target, to obtain a plurality of sampling points;The transmitter Tx position and the characteristic vector of sampling point are input into the deformation network, and the corrected characteristic value is output;Through the radiation network, the corrected characteristic value and the ray direction information are combined to predict the signal strength transmitted by the sampling point to Rx;Based on volume rendering model, the cumulative received signal of all sampling points on each ray to Rx is calculated;With the error of spatial spectrum prediction value and true observation value, the weighted sum of background entropy loss function is taken as the optimization target to train the model.The application combines spatial voxel modeling and double-network decoupling structure, can efficiently fit the channel spatial spectrum distribution under any transmitter position, and performs excellently in multipath complex environment.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and more specifically, to a fast wireless channel modeling method and system based on voxelized radiation fields. Background Technology

[0002] In recent years, with the rapid development of sixth-generation mobile communication systems (6G), communication frequency bands have been continuously expanding to higher frequencies, and application scenarios have become increasingly complex and diverse. Traditional wireless channel modeling methods are facing severe challenges. As a fundamental step in the design, deployment, optimization, and evaluation of communication systems, the accuracy and efficiency of wireless channel modeling directly impact the overall system performance. Therefore, there is an urgent need to develop a new generation of modeling methods with higher accuracy, higher efficiency, and good generalization capabilities.

[0003] Existing wireless channel modeling methods can be mainly divided into three categories: deterministic modeling, statistical modeling, and hybrid modeling methods. Deterministic modeling methods typically rely heavily on environmental modeling, making them difficult to adapt to dynamic and complex scenarios. They also incur high computational costs, making them unsuitable for systems with high real-time requirements. Statistical modeling methods lack detailed characterization of specific scenarios and cannot provide positional information such as angle distributions, thus hindering the achievement of task-oriented, refined modeling. Hybrid modeling combines deterministic and statistical modeling, but these methods still rely on scene geometry information, presenting certain challenges in practical deployment.

[0004] In recent years, with the development of computer vision, Neural Radiation Field (NeRF) has attracted widespread attention as an emerging 3D reconstruction technology. NeRF implicitly represents 3D scenes through a multilayer perceptron (MLP), achieving high-quality viewpoint synthesis with only sparse images. Inspired by this, existing research has attempted to extend NeRF to wireless channel modeling, modeling scenes as wireless radiation fields to generate spatial angular power spectra. However, due to its reliance on deep MLP networks and dense sampling, training time is long and inference speed is slow, making it difficult to meet the needs of real-time applications. Furthermore, some methods require auxiliary equipment such as LiDAR to acquire environmental point cloud data, increasing system complexity and deployment costs. In summary, while existing wireless channel modeling methods based on NeRF have achieved breakthroughs in accuracy, they still have shortcomings in modeling speed, data dependency, and deployment flexibility. Therefore, a new channel modeling framework is urgently needed that can significantly improve modeling efficiency while maintaining high accuracy, reducing dependence on training data and external sensors, and meeting the urgent need for efficient channel modeling in future 6G wireless systems.

[0005] Chinese patent document CN118826933A discloses a method, apparatus, device, and storage medium for predicting wireless channel response. The method includes: receiving input prediction point data, which includes a prediction frequency and prediction point coordinates; acquiring a signal ray set centered on the prediction point coordinates and encoding the signal ray set to obtain a signal direction code, wherein the signal ray set contains at least two virtual signal rays; uniformly arranging sampling voxel points along the virtual signal rays using a preset sampling distance and a preset sampling interval, and encoding the sampling voxel points to obtain a voxel position code; inputting the prediction frequency, voxel position code, and signal direction code into a preset neural network to output voxel density characteristics, voxel signal amplitude, and voxel signal phase; and calculating the wireless channel response at the predicted point coordinates based on the voxel density characteristics, voxel signal amplitude, and voxel signal phase. However, the modeling accuracy in this paper is insufficient to meet different needs, and the modeling efficiency is low. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the purpose of this invention is to provide a fast wireless channel modeling method and system based on voxelized radiation fields.

[0007] A fast wireless channel modeling method based on voxelized radiation field provided by the present invention includes:

[0008] Step S1: Divide the space to be modeled into a three-dimensional voxel mesh, with each voxel storing the volume density and feature vector; obtain multiple sampling points by sampling along the ray from the receiver Rx to the target direction, and obtain the volume density and feature vector of each sampling point through trilinear interpolation;

[0009] Step S2: Input the feature vectors of the transmitter Tx position and the sampling point into the deformation network, and output the corrected feature values; combine the corrected feature values ​​and ray direction information through the radiation network to predict the signal strength emitted from the sampling point to Rx;

[0010] Step S3: Calculate the cumulative received signal of all sampling points on each ray to Rx based on the volume rendering model. The cumulative received signal is obtained by weighted summation, and the weights are determined by the transmittance of the sampling points and the signal termination probability.

[0011] Step S4: Using the error between the spatial spectrum prediction and the actual observation, combined with the weighted sum of the background entropy loss function as the optimization objective, train the model to approximate the real spatial spectrum.

[0012] Preferably, in step S1, the resolution of the three-dimensional voxel grid is progressively increased during the training process: initially a low-resolution grid is used, and then gradually upsampled to the target resolution through interpolation.

[0013] Preferably, in step S1, a spatial skipping mechanism is introduced in the sampling process: if the density value of a sampling point is lower than a set threshold, the sampling point is skipped in training and inference.

[0014] Preferably, in step S2, the transmitter Tx position, sampling point position, and viewing angle are expanded using position coding for high-frequency feature extension. The formula for position coding is:

[0015] E(p)=(sin(2 0 πp),cos(2 0 πp),…,sin(2 L-1 πp),cos(2 L-1 πp))#

[0016] E(·) acts on the input vector element by element, encoding each component independently, and L is a hyperparameter that controls the encoding dimension.

[0017] Preferably, in step S2, the deformation network and the radiation network are lightweight multilayer perceptrons, wherein the deformation network output is used to correct the feature vector to decouple the influence of the static characteristics of the scene and the dynamic position of Tx, and the radiation network outputs the direction-dependent signal intensity.

[0018] Preferably, in step S3, the method for calculating the accumulated received signal is as follows:

[0019]

[0020] α i =1-exp(-σ(x) i )δ i )#

[0021]

[0022] Where K is the number of sampling points along the ray, w i =T i α i It is the weight of the i-th virtual Tx, α i and α j T represents the probability that the signal terminates at the i-th and j-th sampling points, respectively. i From x i The cumulative transmittance to Rx, δ i It is the distance between adjacent sampling points.

[0023] Preferably, in step S4, the background entropy loss function is:

[0024]

[0025] in It is a small batch of ray sets, T K (ω) represents the cumulative transmittance along ray ω;

[0026] The total loss function is:

[0027] L = L spectrum +λ bg L bg

[0028] Where L spectrum Let λ be the mean square error of the spatial spectrum. bg These are the weighting coefficients;

[0029]

[0030] Where |Ω|=M×N represents the angular resolution of the spatial spectrum, corresponding to M azimuth angles φ and N elevation angles θ uniformly sampled on the unit upper hemisphere of Rx.

[0031] Preferably, in step S1, the local attributes encoded by the feature vector include: conductivity and dielectric constant.

[0032] A fast wireless channel modeling system based on voxelized radiation field provided by the present invention includes:

[0033] Module M1: Divides the space to be modeled into a three-dimensional voxel mesh, with each voxel storing the volume density and feature vector; multiple sampling points are obtained by sampling along the ray from the receiver Rx to the target direction, and the volume density and feature vector of each sampling point are obtained by trilinear interpolation;

[0034] Module M2: Inputs the feature vectors of the transmitter Tx position and the sampling point into the deformation network, and outputs the corrected feature values; combines the corrected feature values ​​with the ray direction information through the radiation network to predict the signal strength emitted from the sampling point to Rx;

[0035] Module M3: Calculates the cumulative received signal of all sampling points on each ray to Rx based on the volume rendering model. The cumulative received signal is obtained by weighted summation, and the weights are determined by the transmittance of the sampling points and the signal termination probability.

[0036] Module M4: The model is trained to approximate the real spatial spectrum by using the error between the spatial spectrum prediction and the actual observation as the optimization objective, combined with the weighted sum of the background entropy loss function.

[0037] Preferably, in module M1, the resolution of the three-dimensional voxel grid is progressively increased during training: initially a low-resolution grid is used, and then gradually upsampled to the target resolution through interpolation.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. This invention combines spatial voxel modeling with a dual-network decoupling structure to efficiently fit the channel spatial spectrum distribution at any transmitter location, demonstrating excellent performance in complex multipath environments. Compared to traditional models based on ray tracing or unified function approximation, this invention has significant advantages in modeling accuracy, generalization ability, and training speed, and possesses good potential for engineering implementation.

[0040] 2. This invention provides a fast and efficient wireless channel modeling method that can accurately predict the received spatial spectrum and RSSI even when the transmitter is moving. By voxelizing the radiation field and decoupling the static environmental features from the dynamic Tx effect using a deformation network, it achieves state-of-the-art performance on both RFID and BLE datasets. At the same time, the proposed method has the shortest training time and near real-time inference speed.

[0041] 3. This invention further improves the convergence speed and robustness of the model under sparse measurement conditions by introducing progressive learning, spatial skipping strategy and background entropy loss term. Attached Figure Description

[0042] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0043] Figure 1 This is a schematic diagram illustrating the reflection, diffraction, and scattering of electromagnetic waves with environmental objects during propagation in this invention.

[0044] Figure 2 This is a schematic diagram of virtual transmission point sampling along the receiving direction in this invention;

[0045] Figure 3 This is a schematic diagram illustrating the acquisition of volume density and features using trilinear interpolation of sampling points in this invention;

[0046] Figure 4 This is a structural block diagram of the deformation network and radiation network described in the embodiments of the present invention;

[0047] Figure 5 The method described in the embodiments of the present invention, NeRF, is shown in this invention. 2 A visual comparison of the spatial spectrum reconstruction results between WRF-GS+ and WRF-GS+;

[0048] Figure 6 The present invention shows a comparison of the cumulative distribution function (CDF) of the method described in the embodiments of the present invention with that of existing methods on structural similarity (SSIM);

[0049] Figure 7The present invention shows a performance comparison chart between the method described in the embodiments of the present invention and existing methods under different training set sizes;

[0050] Figure 8 The method described in the embodiments of the present invention and NeRF are shown in this invention. 2 A comparison of RSSI prediction errors on the BLE dataset using MRI. Detailed Implementation

[0051] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0052] Definitions of abbreviations and key terms:

[0053] 6G Sixth Generation Mobile Communications System

[0054] NeRF (Neural Radiance Field)

[0055] Rx Receiver

[0056] Tx Transmitter

[0057] Mean Square Error (MSE)

[0058] WRF Wireless Radiance Field

[0059] GS Gaussian Splatting

[0060] SSIM Structural Similarity Index

[0061] CDF Cumulative Distribution Function

[0062] RSSI Received Signal Strength Indicator

[0063] Bluetooth Low Energy (BLE)

[0064] MRI Model-driven Radio Interpolation Method

[0065] This invention provides a fast wireless channel modeling method, VoxelR, based on voxelized radiation fields. This method is applicable to scenarios with fixed receivers (Rx) and multiple transmitters (Tx), achieving efficient spatial spectrum synthesis through spatial direction sampling and signal accumulation. The method includes the following steps:

[0066] The first step is to obtain the corresponding spatial angular power spectrum for each transmitter through ray sampling. Rays from the receiver to a specified direction are sampled uniformly to form multiple sampling points. Each sampling point is considered a virtual emission source, and its spatial position is determined by the receiver position, the ray direction, and the sampling distance.

[0067] The second step involves calculating the volume density and its radiated signal intensity at each sampling point. Based on the volume rendering model, the transmittance and attenuation factor of the ray sampling points are calculated, and the signal contribution of each sampling point to the receiver is accumulated. Signal accumulation uses a weighted summation method, with the weights determined by the transmittance and termination probability.

[0068] The third step is to synthesize the accumulated signals along all ray directions to obtain the overall received signal at the receiver. This signal is then used to calculate the spatial spectrum distribution, enabling the modeling of the spatial characteristics of the wireless channel.

[0069] The fourth step is to train the model to approximate the real spatial spectrum by using the error between the predicted and actual observed values ​​of the spatial spectrum as the optimization objective.

[0070] This invention proposes an explicit-implicit hybrid representation method for wireless signal radiation fields based on voxel grids, which effectively improves the training and inference speed of channel modeling.

[0071] In one specific implementation, in the first step, the space to be modeled is divided into a fixed-resolution three-dimensional voxel mesh. Each voxel stores a density value to represent the probability that the location is a valid signal source and a feature vector to represent local material or propagation environment properties. For each sampling point on a propagation ray, its density and features are obtained through trilinear interpolation. The model initially uses a coarser-resolution voxel mesh to represent the signal field, and then gradually increases the mesh resolution during training, upsampling the original voxels each time through interpolation until the target accuracy is reached. This strategy reduces the risk of overfitting and accelerates convergence. Furthermore, a spatial skipping mechanism is introduced during sampling. If the density value of a sampling point is lower than a set threshold, that point is skipped in subsequent training and inference to reduce redundant computation during training and inference, saving resources.

[0072] In the second step, to improve the model's ability to model changes in the transmitter's Tx position, two lightweight MLPs are introduced as a deformation network and a radiation network. The deformation network takes the Tx position and feature vector as input and outputs a corrected feature value, effectively decoupling the static characteristics of the scene from the radiation field changes caused by the dynamic position of Tx, thus improving the ability to generalize to the Tx position. The radiation network combines the corrected features and direction information to predict the signal strength transmitted from that point to Rx. Furthermore, positional encoding technology is used to perform high-frequency feature extension on the transmitter position, sampling point position, and ray direction to capture the rapid spatial changes in the wireless signal.

[0073] In the fourth step, a background entropy loss function is introduced to help the model make clearer judgments about ray occlusion and accelerate convergence. The final training objective is to minimize the weighted sum of the spatial spectrum error and the background entropy loss.

[0074] The embodiments of the present invention will be described in further detail below.

[0075] The channel modeling method described in this invention is suitable for scenarios with high requirements for deployment speed and real-time performance. For example... Figure 1 As shown, electromagnetic waves undergo various interactions on the surface of an object during propagation, including reflection, diffraction, and scattering, making these surfaces new sources of signal radiation. Therefore, the receiver can receive signals from multiple directions. To calculate the signal received by Rx from a specific direction ω0, we sample along that direction and treat each sampling point as a virtual transmitter, such as... Figure 2 As shown. We define the position x of the i-th virtual Tx. i for

[0076] x i =P Rx +r i ω0#(1)

[0077] Where P Rx Indicates the location of the receiving end, r i It is the distance from Rx to the i-th sampling point.

[0078] We represent the radiation field using two voxel grids: a density voxel grid and a density voxel grid. and feature voxel grid Among them (L) x ,L y ,L z ) represents the size of the voxel grid, and F represents the dimension of the feature space. For each sample X i ,like Figure 3 As shown, we obtain its volume density σ(X) through trilinear interpolation. i ) and feature vector Feat(X) i ):

[0079]

[0080] Where the volume density σ(x) i The feature Feat(x) represents the probability that this point is a virtual Tx due to the propagation effect. i It depends only on spatial location and encodes physical parameters such as conductivity and dielectric constant.

[0081] like Figure 4 As shown, we employ two shallow MLPs to predict the direction-dependent signal radiated from each sampling point. A deformation network D is used to model the influence of the Tx location on local multipath characteristics and outputs the deformation feature ΔFeat(x). i This allows for the modeling of the radiation field related to the location Tx. Subsequently, the deformation features and the original features are fused, and the radiation signal S(x) is predicted by the radiation network R. i This method improves the model's generalization ability by introducing a deformable network D to decouple the dynamic and static radiation fields caused by the movement of the Tx position. Furthermore, we adjust the Tx position P... T Sampling point location x i And the viewpoint direction -ω0 applies position encoding to capture high-frequency changes:

[0082] E(p)=(sin(2 0 πp),cos(2 0 πp),…,sin(2 L-1 πp),cos(2 L-1 πp))#(4)

[0083] E(·) acts on the input vector element by element, encoding each component independently, and L is a hyperparameter that controls the encoding dimension.

[0084] For each direction ω0, the cumulative received signal R(ω0) is calculated using the following body rendering formula:

[0085]

[0086] α i =1-exp(-σ(x) i )δ i )#(6)

[0087]

[0088] Where K is the number of sampling points along the ray, w i =T i α i It is the weight of the i-th virtual Tx, α i and α jT represents the probability that the signal terminates at the i-th and j-th sampling points, respectively. i From x i The cumulative transmittance to Rx, δ i This is the distance between adjacent sampling points. For the spatial spectrum prediction task, the training objective is to minimize the mean square error (MSE) of the signal prediction error in each direction:

[0089]

[0090] Where |Ω|=M×N represents the angular resolution of the spatial spectrum, corresponding to M azimuth angles φ and N elevation angles θ uniformly sampled on the unit upper hemisphere of Rx.

[0091] We employ the following three training strategies to optimize performance and accelerate training convergence and inference speed.

[0092] First, a progressive learning strategy is adopted. In the initial stage of training, a higher resolution voxel grid is used, with a voxel count of [missing value]. The mesh resolution is progressively increased over M stages. In each stage, the number of voxels is doubled, and V is adjusted using trilinear interpolation. density and V feature Upsampling is performed until the final resolution (L) is reached. x ,L y ,L z This strategy can reduce the risk of the algorithm getting stuck in local optima and improve the stability of training.

[0093] Furthermore, since most of the space in a real wireless propagation environment is air, contributing very little to the received signal, a uniform sampling strategy would result in a large amount of invalid computation. Therefore, to improve sampling efficiency, we introduce a spatial skipping strategy during the training and inference phases. If the volume density of a sampling point is lower than a set threshold τ, indicating that it is located in air, then that point is skipped during subsequent training and inference processes.

[0094] Finally, we introduce the background entropy loss function L bb This is used to improve the ability to identify occlusion situations.

[0095]

[0096] in It is a small batch of ray sets, T K (ω) represents the cumulative transmittance along ray ω. This loss prompts the network to make more precise predictions about occluded areas, thus accelerating training convergence. The final total loss function is...

[0097] L = L spectrum +λ bg L bg

[0098] Where L spectrum From formula (8), λ bg These are the weighting coefficients.

[0099] To verify the performance of the voxelized radiation field-based approach proposed in this invention in real-world scenarios, we conducted simulation experiments using publicly available wireless channel datasets. The experiments covered multiple aspects, including spatial spectrum reconstruction performance, image structure similarity, data scale robustness, training and inference efficiency, and Received Signal Strength Indication (RSSI) prediction, as detailed below:

[0100] The experiment used NeRF 2 This is an open-source RFID dataset. The dataset was collected in an indoor experimental environment measuring 23m × 24m × 2m. The receivers were fixed, equipped with a 4×4 antenna array operating at 915MHz. The transmitters were movable RFID tags randomly placed in space and continuously transmitting RN16 signals. The dataset contains 6,123 transmitter locations, each corresponding to a 360×90 spatial spectrum, representing the distribution of received signal strength in all directions at that location. Unless otherwise specified, by default, 80% of the data is used for training and 20% for testing.

[0101] To verify the effectiveness of the method of the present invention, it is compared with the following two existing methods, namely NeRF, which is the first to use neural radiation fields for wireless channel modeling. 2 And WRF-GS+, a wireless radiation field (WRF) modeling method based on 3D Gaussian sputtering (GS).

[0102] Figure 5 The method of this invention is combined with NeRF 2 The spatial spectra synthesized by WRF-GS+ at multiple transmitter locations are visualized and compared with the real spectra. It is evident that the spatial spectrum generated by the method proposed in this invention is closest to the real image, with a clearer transition between high-intensity and low-intensity regions, indicating that it better preserves high-frequency information. This demonstrates that the method of this invention can more accurately capture rapid spatial fluctuations caused by small-scale fading effects. In contrast, although WRF-GS+ is comparable to NeRF... 2 It performs better, but because it uses an inherently smooth and continuous Gaussian volume, the resulting image has blurred edges, especially in the boundary areas, and loses sharp details.

[0103] Figure 6The Structural Similarity Index (SSIM) was further used to evaluate the similarity between the synthesized image and the real spatial spectral map. Cumulative Distribution Function (CDF) results showed that the median SSIM of the method in this invention was 0.9035, which is superior to WRF-GS+ (0.8813) and NeRF. 2 (0.7991). This invention enhances the modeling capability for changes at a moving transmitter by introducing a deformable network to separate the dynamic and static fields, thereby significantly improving the performance of spatial spectrum synthesis. In contrast, the original NeRF is used for static radiation field modeling, and therefore has limited ability to sense changes in the radiation field caused by changes in the Tx position.

[0104] Figure 7 The training data was set to five proportions: 20%, 40%, 60%, 80%, and 100%, and the performance of each method was compared under different training set sizes. The results show that, under all data sizes, the proposed method outperforms the comparative methods, demonstrating good robustness to sparse data. Furthermore, the model tends to converge when using approximately 60% of the training data, reflecting its excellent data efficiency. This is because, compared to NeRF... 2 Compared to purely implicit representations, this invention uses explicit voxel grids to represent the radiation field and ensures local continuity through interpolation, thereby playing a regularization role when the data is sparse and reducing the risk of overfitting.

[0105] We further compared the training and inference times of each method. NeRF 2 The training time was 15 hours and 30 minutes, WRF-GS+ took 2 hours and 40 minutes, while the algorithm proposed in this invention only takes 20 minutes, which is 8 times faster than WRF-GS+ and faster than NeRF. 2 Nearly 47 times faster. During the inference phase, the inference time of this invention is 0.042 seconds, approximately the same as NeRF. 2 It is 1 / 10 of the time (0.396 seconds), close to WRF-GS+ (0.01 seconds). This is thanks to VoxelRF using trilinear interpolation of voxel meshes instead of NeRF. 2 The proposed method utilizes a deep MLP network and skips sampling points with low volume density. These results demonstrate that the method is well-suited for deployment in latency-sensitive or resource-constrained scenarios.

[0106] Considering that the RSSI value can be obtained from the spatial spectrum by integrating the directional power, further in NeRF 2 The performance of this method in RSSI prediction was tested on the open-source BLE dataset. The BLE dataset consisted of 21 receivers and one mobile transmitter deployed in a 15,000 square foot indoor scene, containing 2,000 valid RSSI measurements (greater than -100 dBm). Results are as follows... Figure 8As shown, the RSSI prediction median error of the proposed solution is 2.66 dB, which is superior to NeRF. 2 (2.85dB) and the model-driven wireless interpolation method MRI (6.74dB). Overall, the combined algorithm proposed in this invention performs better.

[0107] The present invention also provides a fast wireless channel modeling system based on voxelized radiation fields. The fast wireless channel modeling system based on voxelized radiation fields can be implemented by executing the process steps of the fast wireless channel modeling method based on voxelized radiation fields. That is, those skilled in the art can understand the fast wireless channel modeling method based on voxelized radiation fields as a preferred embodiment of the fast wireless channel modeling system based on voxelized radiation fields.

[0108] A fast wireless channel modeling system based on voxelized radiation field, comprising:

[0109] Module M1: Divides the space to be modeled into a three-dimensional voxel mesh, with each voxel storing the volume density and feature vector; multiple sampling points are obtained by sampling along the ray from the receiver Rx to the target direction, and the volume density and feature vector of each sampling point are obtained by trilinear interpolation;

[0110] Module M2: Inputs the feature vectors of the transmitter Tx position and the sampling point into the deformation network, and outputs the corrected feature values; combines the corrected feature values ​​with the ray direction information through the radiation network to predict the signal strength emitted from the sampling point to Rx;

[0111] Module M3: Calculates the cumulative received signal of all sampling points on each ray to Rx based on the volume rendering model. The cumulative received signal is obtained by weighted summation, and the weights are determined by the transmittance of the sampling points and the signal termination probability.

[0112] Module M4: The model is trained to approximate the real spatial spectrum by using the error between the spatial spectrum prediction and the actual observation as the optimization objective, combined with the weighted sum of the background entropy loss function.

[0113] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0114] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A fast wireless channel modeling method based on voxelized radiation fields, characterized in that, include: Step S1: Divide the space to be modeled into a three-dimensional voxel mesh, with each voxel storing the volume density and feature vector; Multiple sampling points are obtained by sampling the ray along the receiver Rx to the target direction, and the volume density and feature vector of each sampling point are obtained by trilinear interpolation; Step S2: Input the feature vectors of the transmitter Tx position and the sampling points into the deformation network, and output the corrected feature values; combine the corrected feature values ​​and ray direction information through the radiation network to predict the signal intensity emitted by each sampling point toward Rx; Step S3: Calculate the cumulative received signal of all sampling points on each ray to Rx based on the volume rendering model. The cumulative received signal is obtained by weighted summation, and the weights are determined by the transmittance of the sampling points and the signal termination probability. Step S4: Using the error between the spatial spectrum prediction and the actual observation, combined with the weighted sum of the background entropy loss function as the optimization objective, train the model to approximate the real spatial spectrum.

2. The fast wireless channel modeling method based on voxelized radiation field according to claim 1, characterized in that, In step S1, the resolution of the three-dimensional voxel grid is progressively increased during the training process: initially a low-resolution grid is used, and then it is gradually upsampled to the target resolution through interpolation.

3. The fast wireless channel modeling method based on voxelized radiation field according to claim 1, characterized in that, In step S1, a spatial skipping mechanism is introduced in the sampling process: if the density value of a sampling point is lower than a set threshold, the sampling point is skipped during training and inference.

4. The fast wireless channel modeling method based on voxelized radiation field according to claim 1, characterized in that, In step S2, high-frequency feature expansion is performed using position coding for the transmitter Tx position, sampling point position, and viewing direction. The formula for the position coding is: in Applying element-wise to the input vector, each component Independent encoding, Hyperparameters for controlling the encoding dimension.

5. The fast wireless channel modeling method based on voxelized radiation field according to claim 1, characterized in that, In step S2, the deformation network and the radiation network are lightweight multilayer perceptrons, wherein the deformation network output is used to correct feature values ​​to decouple the influence of static characteristics of the scene and dynamic position of Tx, and the radiation network outputs direction-dependent signal strength.

6. The fast wireless channel modeling method based on voxelized radiation field according to claim 1, characterized in that, In step S3, the method for calculating the accumulated received signal is as follows: in It is the number of sampling points along the ray. The direction of view; Location of each sampling point; It is the first The weight of a virtual Tx, For the radiation network Predicting radiation signals; and The signals are at the 1st and 2nd respectively. The and the first The probability of terminating at each sampling point. Bulk density; From The cumulative transmittance to Rx, It is the distance between adjacent sampling points.

7. The fast wireless channel modeling method based on voxelized radiation field according to claim 1, characterized in that, In step S4, the background entropy loss function is: in It is a small batch of ray collections. Indicates along the ray The cumulative transmittance.

8. The fast wireless channel modeling method based on voxelized radiation field according to claim 1, characterized in that, In step S1, the local attributes encoded by the feature vector include conductivity and dielectric constant.

9. A fast wireless channel modeling system based on voxelized radiation fields, characterized in that, include: Module M1: Divides the space to be modeled into a three-dimensional voxel mesh, with each voxel storing the volume density and feature vector; Multiple sampling points are obtained by sampling the ray along the receiver Rx to the target direction, and the volume density and feature vector of each sampling point are obtained by trilinear interpolation; Module M2: Inputs the feature vectors of the transmitter Tx position and the sampling points into the deformation network, and outputs the corrected feature values; combines the corrected feature values ​​with the ray direction information through the radiation network to predict the signal strength emitted by each sampling point toward Rx; Module M3: Calculates the cumulative received signal of all sampling points on each ray to Rx based on the volume rendering model. The cumulative received signal is obtained by weighted summation, and the weights are determined by the transmittance of the sampling points and the signal termination probability. Module M4: The model is trained to approximate the real spatial spectrum by using the error between the spatial spectrum prediction and the actual observation as the optimization objective, combined with the weighted sum of the background entropy loss function.

10. The fast wireless channel modeling system based on voxelized radiation field according to claim 9, characterized in that, In module M1, the resolution of the three-dimensional voxel grid is progressively increased during training: initially a low-resolution grid is used, and then gradually upsampled to the target resolution through interpolation.

Images