Channel reconstruction method and device, computer equipment and storage medium

Abstract

The invention provides a channel reconstruction method and device, computer equipment and a storage medium. The channel reconstruction method comprises the following steps: constructing a first feature vector; distributing the feature information of the plurality of obstacles to a plurality of feature buckets according to the hash value of the feature information of each obstacle in the first feature vector; calculating an attention result of the feature information of the plurality of target obstacles in the same feature bucket or adjacent feature buckets; splicing all attention results to obtain a second feature vector; respectively decomposing the second feature vector along a height axis and a width axis to obtain a plurality of first sub-feature vectors and a plurality of second sub-feature vectors; splicing the query vector and the first attention result of the plurality of first sub-feature vectors and the query vector and the second attention result of the plurality of second sub-feature vectors to obtain a third feature vector; and estimating a key channel parameter through the third feature vector. According to the method and the device, the estimation accuracy of the channel characteristics can be improved on the basis of reducing the calculation complexity.

Description

Technical Field

[0001] This application relates to the field of wireless communication technology, specifically to a channel reconstruction method, apparatus, computer device, and storage medium. Background Technology

[0002] With the advent of the 6G era, the migration of communication spectrum to higher frequency bands such as millimeter waves and terahertz has become an inevitable trend. The enormous bandwidth contained in these frequency bands is key to achieving revolutionary application experiences such as ultra-high speed and ultra-low latency. However, opportunities and challenges coexist. The "light-like" propagation physical characteristics of high-frequency signals make them more dependent on the propagation environment than ever before. In complex indoor environments, the movement of walls, furniture, equipment, and even people can have a decisive impact on the signal propagation path, forming a dynamic and complex channel environment strongly coupled with the physical layout. Therefore, if the channel characteristics under a specific physical layout can be accurately estimated in advance, and the entire channel can be reconstructed, it will greatly ensure the optimization of resource allocation and the guarantee of user service quality.

[0003] Current techniques for channel feature acquisition mainly include deterministic models and statistical models. Deterministic models require enormous computational resources, and their results are extremely sensitive to the accuracy of the model. Even a slight error in the placement of an object or a small deviation in a wall material parameter can lead to significant errors in channel modeling and reconstruction. Statistical models, on the other hand, extract the distribution patterns of channel parameters in a typical environment through statistical analysis of a large amount of historical measurement data. However, they ignore specific characteristics, resulting in coarse estimation results with significant limitations. Summary of the Invention

[0004] In view of this, this application proposes a channel reconstruction method, apparatus, computer device, and storage medium.

[0005] A first aspect of this application provides a channel reconstruction method, the method comprising: A first feature vector is constructed, which includes feature information of multiple obstacles in the indoor environment; the feature information includes shape information, position information, and material information. Based on the hash value of the feature information of each obstacle in the first feature vector, the feature information of the multiple obstacles is allocated to multiple feature buckets; For multiple target obstacles within the same feature bucket or adjacent feature buckets, calculate the attention results of the feature information of the multiple target obstacles; The second feature vector is obtained by concatenating all the attention results corresponding to the multiple feature buckets. The second feature vector is decomposed along the height axis and the width axis to obtain multiple first sub-feature vectors and multiple second sub-feature vectors; each first sub-feature vector is used to characterize the feature information of the multiple obstacles in the same feature dimension; each second sub-feature vector is used to characterize the feature information of the corresponding obstacle in multiple feature dimensions; The query vector is concatenated with the first attention result of the plurality of first sub-feature vectors and the second attention result of the query vector with the plurality of second sub-feature vectors to obtain a third feature vector; the query vector represents the parameter estimation task and is used to determine the feature information most relevant to channel parameter estimation from the second feature vector; The key channel parameters, including path loss, time of arrival, angle of arrival, and angle of departure, are estimated using the third feature vector.

[0006] This embodiment of the application allocates the feature information of multiple obstacles into multiple feature buckets based on the hash value of the feature information of each obstacle in the first feature vector. For multiple target obstacles in the same feature bucket or adjacent feature buckets, the attention result of the feature information of multiple target obstacles is calculated, and all attention results corresponding to multiple feature buckets are concatenated to obtain the second feature vector. This can significantly reduce computational complexity and thus help improve the estimation efficiency of channel features. Preferably, this embodiment of the application further decomposes the second feature vector along the height axis and the width axis to obtain multiple first sub-feature vectors and multiple second sub-feature vectors, respectively. The query vector is concatenated with the first attention result of the multiple first sub-feature vectors and the second attention result of the query vector with the multiple second sub-feature vectors to obtain the third feature vector. This can establish the interaction relationship between the feature information of multiple obstacles in the indoor environment and the channel parameters along different dimensions, thereby greatly improving the estimation accuracy of channel features.

[0007] In this embodiment of the application, constructing the first feature vector includes: For any one of the plurality of obstacles, obtain the shape information, position information, and material information of the obstacle; Determine the transmitter location information and receiver location information; The shape information, the position information, the material information, the transmitter position information, and the receiver position information are used as the feature information of the obstacle; The first feature vector is constructed based on the feature information of the multiple obstacles.

[0008] In this embodiment of the application, the feature information of the plurality of obstacles is allocated to a plurality of feature buckets according to the hash value of the feature information of each obstacle in the first feature vector, including: The hash value of the feature information of each obstacle is calculated using a hash function; For any two obstacles among the plurality of obstacles, if the hash values ​​of the two obstacles are the same, then the feature information of the two obstacles is assigned to the same feature bucket; If the difference between the hash values ​​of the two obstacles is equal to a preset difference threshold, then the feature information of the two obstacles will be assigned to adjacent feature buckets.

[0009] In this embodiment of the application, after decomposing the second feature vector along the height axis and the width axis to obtain multiple first sub-feature vectors and multiple second sub-feature vectors respectively, the method further includes: For any one of the plurality of first sub-feature vectors, calculate the first relevance score between the query vector and the first sub-feature vector; The multiple first correlation scores corresponding to the multiple first sub-feature vectors are normalized to obtain multiple first probability values; the sum of the multiple first probability values ​​is 1; The first attention result is obtained by weighted summation of the multiple first sub-feature vectors using the multiple first probability values; For any one of the plurality of second sub-feature vectors, calculate a second relevance score between the query vector and the second sub-feature vector; The multiple second correlation scores corresponding to the multiple second sub-feature vectors are normalized to obtain multiple second probability values; the sum of the multiple second probability values ​​is 1; The second attention result is obtained by weighted summation of the multiple second sub-feature vectors using the multiple second probability values.

[0010] An embodiment of the second aspect of this application provides a channel reconstruction model, including: an encoder and a decoder; The encoder is used to acquire a first feature vector, which includes feature information of multiple obstacles in the indoor environment, transmitter position information, and receiver position information; the feature information includes shape information, position information, and material information; based on the hash value of the feature information of each obstacle in the first feature vector, the feature information of the multiple obstacles is assigned to multiple feature buckets; for multiple target obstacles in the same feature bucket or adjacent feature buckets, the attention result of the feature information of the multiple target obstacles is calculated; and all attention results corresponding to the multiple feature buckets are concatenated to obtain a second feature vector; The decoder is used to decompose the second feature vector along the height axis and the width axis to obtain multiple first sub-feature vectors and multiple second sub-feature vectors respectively; each first sub-feature vector is used to characterize the feature information of the multiple obstacles in the same feature dimension; each second sub-feature vector is used to characterize the feature information of the corresponding obstacle in multiple feature dimensions; the query vector is concatenated with the first attention result of the multiple first sub-feature vectors and the second attention result of the query vector with the multiple second sub-feature vectors to obtain a third feature vector; the query vector represents the parameter estimation task and is used to determine the feature information most relevant to channel parameter estimation from the second feature vector; the third feature vector is used to estimate key channel parameters, including path loss, time of arrival, angle of arrival, and departure angle.

[0011] In this embodiment of the application, the channel reconstruction model is trained through the following steps: Construct a multi-task loss function, which includes estimated losses for path loss, arrival time, arrival angle, and departure angle, as well as loss weights for each estimated loss. The channel reconstruction model is trained iteratively through the multi-task loss function until the difference between the key channel parameters output by the decoder and the target label converges to within a preset threshold. The target labels include real labels and pseudo labels; the pseudo labels are generated based on environmental data and ray tracing parameters through a simulated signal propagation path.

[0012] In this embodiment of the application, the model training further includes: By using cosine annealing to automatically reduce the learning rate during training, the channel reconstruction model can obtain a better local minimum during the convergence phase, thereby improving the joint estimation accuracy of path loss, arrival time, arrival angle, and departure angle.

[0013] An embodiment of the third aspect of this application provides a channel reconstruction apparatus, comprising: The first feature vector construction module is used to construct a first feature vector, which includes feature information of multiple obstacles in the indoor environment, transmitter location information, and receiver location information; the feature information includes shape information, location information, and material information. The feature information allocation module is used to allocate the feature information of the multiple obstacles to multiple feature buckets according to the hash value of the feature information of each obstacle in the first feature vector; The attention result calculation module is used to calculate the attention result of the feature information of multiple target obstacles in the same feature bucket or adjacent feature buckets; The second feature vector generation module is used to concatenate all attention results corresponding to the multiple feature buckets to obtain the second feature vector; The second feature vector decomposition module is used to decompose the second feature vector along the height axis and the width axis to obtain multiple first sub-feature vectors and multiple second sub-feature vectors respectively; each first sub-feature vector is used to characterize the feature information of the multiple obstacles in the same feature dimension; each second sub-feature vector is used to characterize the feature information of the corresponding obstacle in multiple feature dimensions; The third feature vector generation module is used to concatenate the query vector with the first attention results of the plurality of first sub-feature vectors and the second attention results of the query vector with the plurality of second sub-feature vectors to obtain a third feature vector; the query vector represents the parameter estimation task and is used to determine the feature information most relevant to channel parameter estimation from the second feature vector; The critical channel parameter estimation module is used to estimate critical channel parameters through the third feature vector, wherein the critical channel parameters include path loss, time of arrival, angle of arrival, and departure angle.

[0014] An embodiment of the fourth aspect of this application provides a computer device including a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the channel reconstruction method described in the first aspect above.

[0015] An embodiment of the fifth aspect of this application provides a computer-readable storage medium storing computer instructions for causing a computer to perform the channel reconstruction method described in the first aspect above.

[0016] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0017] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A schematic flowchart of a channel reconstruction method provided in an embodiment of this application is shown; Figure 2 A schematic diagram of the channel reconstruction model provided in one embodiment of this application is shown; Figure 3 This invention provides a schematic diagram of the structure of a channel reconstruction apparatus according to an embodiment of the present application. Figure 4 This illustration shows a schematic diagram of the structure of a computer device according to an embodiment of this application; Figure 5 A schematic diagram of a storage medium provided in one embodiment of this application is shown. Detailed Implementation

[0018] Exemplary embodiments of this application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of this application and to fully convey the scope of this application to those skilled in the art.

[0019] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains.

[0020] The technical scenarios involved in the embodiments of this application are described below.

[0021] Current techniques for channel feature acquisition exhibit a clear polarization. The first approach prioritizes high physical fidelity at the expense of efficiency and scalability. Deterministic modeling, exemplified by ray tracing, is the most representative. Its principle involves simulating the propagation paths of thousands of electromagnetic waves in a meticulously constructed 3D environment based on geometric optics and uniform diffraction theory, thereby achieving subsequent feature modeling. Theoretically, it can provide highly accurate channel estimates; however, it requires constructing a 3D indoor model accurate to the centimeter level with complete material information. Furthermore, the simulation process involves massive geometric intersections and physical optics calculations, demanding enormous computational resources, with single simulations taking anywhere from minutes to hours. The results are extremely sensitive to the model's accuracy; even a slight error in object placement or a minute deviation in a wall material parameter can lead to significant errors in channel modeling and reconstruction. While field measurement methods can provide accurate data, their application limitations are equally significant. The expensive specialized channel measurement equipment, cumbersome on-site deployment and calibration procedures, and lengthy measurement cycles limit its use to laboratory verification or sampling tests in a very few critical scenarios. Conducting comprehensive field measurements for every potential deployment scenario is impractical in terms of both economy and time; it is essentially a one-off, point-based solution rather than a universal modeling tool. The second type of approach prioritizes computational efficiency and versatility at the expense of location-specificity and accuracy. Represented by statistical channel models defined by standardization organizations such as 3GPP, these methods extract the distribution patterns of channel parameters in a typical environment through statistical analysis of large amounts of historical measurement data. Their advantage lies in their simplicity and ease of use. However, their fundamental flaw lies in "generalization" rather than "specialization." While they can well describe average characteristics, they completely ignore specific features. For a particular room, the deterministic impact of the specific location and material of a table or a glass partition is something these models cannot capture. Therefore, the parameter estimates they provide are coarse and probabilistic, with significant limitations.

[0022] In summary, existing technologies suffer from the following problems in channel parameter estimation and channel reconstruction: deterministic models often imply high costs and extremely low efficiency, while statistical models suffer from severe lack of accuracy and loss of environmental details. Therefore, the core technical problem this application aims to solve is: how to construct a fast, accurate, and environmentally aware channel reconstruction model. This model can take specific indoor environmental physical characteristics (such as transceiver coordinates, distance, horizontal difference, obstacle geometry and material information) as input, and through a data-driven intelligent paradigm, bypass complex electromagnetic wave propagation physical calculations to directly and efficiently map and output key channel parameters under this specific layout. This enables accurate estimation and channel reconstruction of indoor high-frequency channel characteristics, providing strong technical support for the intelligent deployment and optimization of 6G networks.

[0023] According to an embodiment of this application, a channel reconfiguration method embodiment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0024] This embodiment provides a channel reconstruction method. Figure 1 This is a flowchart of a channel reconstruction method according to an embodiment of this application, such as... Figure 1 As shown, the process includes the following steps: Step S101: Construct the first feature vector.

[0025] Specifically, the first feature vector includes feature information of multiple obstacles in the indoor environment; the feature information includes shape information, location information, and material information. The shape information includes the geometry of the obstacle, such as the length, width, and height of a rectangular wall; the location information can be represented using a three-dimensional coordinate system, such as the coordinates of the obstacle's center point; the material information includes, but is not limited to, the dielectric constant of concrete, the reflectivity of metal, and the absorptivity of wood.

[0026] In some specific embodiments, step S101 includes steps S1011-S1014: Step S1011: For any one of the plurality of obstacles, obtain the shape information, position information and material information of the obstacle.

[0027] Specifically, the shape, location, and material information of obstacles can be obtained in the following ways: A laser scanner is used to scan the entire indoor space, acquiring shape and location information for each obstacle, which can be presented as a point cloud file. An electromagnetic wave reflection tester is used to measure material properties: for each obstacle, electromagnetic waves of a known frequency are emitted, and the reflection and absorption intensities are measured and labeled according to the IEEE standard material property table.

[0028] Step S1012: Determine the transmitter location information and the receiver location information.

[0029] Step S1013: The shape information, the position information, the material information, the transmitter position information, and the receiver position information are used as the feature information of the obstacle.

[0030] Specifically, the shape information of the obstacle can be represented as a vector. Location information is Material information Launcher location and receiver location .

[0031] Step S1014: Construct the first feature vector based on the feature information of the multiple obstacles.

[0032] Specifically, for multiple obstacles in an indoor environment, the feature information of the multiple obstacles can be combined into a matrix form, i.e. And define this matrix form as the first eigenvector.

[0033] In some specific embodiments, before step S102, the method further includes: scaling the first feature vector v using a Min-Max normalization formula, the scaling formula of which is as follows:

[0034] in, and The minimum and maximum values ​​of this feature across all data are mapped to the interval [0,1], forming a high-dimensional feature tensor E. Missing values ​​are filled using the nearest neighbor interpolation algorithm: for a missing value i, find the nearest known point j and fill i with the value of j, ensuring the tensor is complete. The formula is:

[0035] In some specific embodiments, after scaling the first feature vector v, the method further includes: using the scaled high-dimensional feature tensor E as the first feature vector and inputting it into the encoder of the channel reconstruction model, the encoder being used to perform the following steps S102-S104.

[0036] In some specific embodiments, before step S102, the method further includes: initially extracting local features of the first feature vector through a convolutional layer, that is: using a 2D convolutional kernel (setting the kernel size to 3x3, stride to 1) to process the high-dimensional feature tensor. Applying convolution operations:

[0037] Where W is the weight matrix, b is the bias, and the output is the extracted features. To capture local spatial dependencies, such as interactions between adjacent obstacles.

[0038] Extracted features As the first feature vector, perform the following steps S102.

[0039] Step S102: Based on the hash value of the feature information of each obstacle in the first feature vector, the feature information of the multiple obstacles is allocated to multiple feature buckets.

[0040] In some specific embodiments, step S102 above includes steps S1021-S1023: Step S1021: Calculate the hash value of the feature information of each obstacle using a hash function.

[0041] Step S1022: For any two obstacles among the plurality of obstacles, if the hash values ​​of the two obstacles are the same, the feature information of the two obstacles is assigned to the same feature bucket.

[0042] Step S1023: If the difference between the hash values ​​of the two obstacles is a preset difference threshold, then the feature information of the two obstacles is assigned to adjacent feature buckets.

[0043] In steps S1021-S1023 above, the hash value can be calculated using the following hash function:

[0044] in, The hash value representing the feature information can also be understood as the bucket number of the feature bucket assigned to the feature information of obstacle x; a and b are random vectors, for example a=[3,5,2,4], a=[6], p is a prime number, which can be understood as the total number of the feature buckets, for example: p=7 means there are 7 feature buckets, i.e. 0...6; Here's an example to illustrate how the hash value of the feature information for each obstacle is calculated: When the feature information of obstacle x1 is [1,0,2,3], Therefore, the feature information of obstacle x1 is placed in feature bucket number 4.

[0045] In some specific embodiments, the preset difference threshold can be set to 1. That is, when the bucket numbers of the feature buckets assigned to the feature information of two obstacles differ by 1, they can be determined to be adjacent feature buckets. For example, if the bucket number of the feature bucket assigned to the feature information of obstacle x1 is 1, then the bucket numbers of the adjacent feature buckets are 0 and 2.

[0046] Step S103: For multiple target obstacles in the same feature bucket or adjacent feature buckets, calculate the attention result of the feature information of the multiple target obstacles.

[0047] Specifically, the attention result for calculating the feature information of multiple target obstacles includes the following steps: First, calculate the query vector Q, key vector K, and value vector V for each target obstacle using the following formulas:

[0048]

[0049]

[0050] in, , , These are learnable linear transformation matrices that map to a low-dimensional hash space (dimension). ).

[0051] Secondly, the attention score between the feature information of any two target obstacles is calculated using the following formula:

[0052] in, Indicates target obstacle i With target obstacles j Attention scores between feature information Represents the vector dot product. Indicates the scaling factor; Third, the attention weight matrix is ​​obtained by normalizing all attention scores corresponding to multiple target obstacles using the following formula:

[0053]

[0054] in, Indicates target obstacle i The query vector for the target obstacle j Attention weights of the key vectors, Indicates target obstacle iFor target obstacles j Attention scores between Indicates the number of multiple target obstacles. Indicates the first i The sum of the exponents of all the fractions is normalized so that the sum of the weights is 1.

[0055] Fourth, attention weights are used to perform a weighted summation of the value vectors of each target obstacle to obtain a new representation of the feature information of each target obstacle:

[0056] in, Indicates target obstacle i A new representation of feature information, Indicates target obstacle i The query vector for the target obstacle j Attention weights of the key vectors, Indicates target obstacle j The value vector.

[0057] Finally, the set of new representations of the feature information of multiple target obstacles is used as the attention result of the feature information of multiple target obstacles.

[0058] Step S104: Concatenate all attention results corresponding to the multiple feature buckets to obtain the second feature vector.

[0059] Step S105: Decompose the second feature vector along the height axis and the width axis to obtain multiple first sub-feature vectors and multiple second sub-feature vectors respectively.

[0060] Specifically, the second feature vector is presented in matrix form. Each row of the matrix represents the shape information, position information, material information, transmitter position information, and receiver position information of the corresponding obstacle. Each column represents the feature information of multiple obstacles in the same feature dimension, such as the position information of multiple obstacles.

[0061] More specifically, each first sub-feature vector is used to characterize the feature information of the multiple obstacles under the same feature dimension; each second sub-feature vector is used to characterize the feature information of the corresponding obstacle under multiple feature dimensions.

[0062] In some specific embodiments, after step S105, the method further includes: Step a1: For any one of the plurality of first sub-feature vectors, calculate the first relevance score between the query vector and the first sub-feature vector; Step a2: Normalize the multiple first correlation scores corresponding to the multiple first sub-feature vectors to obtain multiple first probability values; the sum of the multiple first probability values ​​is 1; Step a3: The first attention result is obtained by weighted summation of the multiple first sub-feature vectors using the multiple first probability values.

[0063] In steps a1-a3 above, attention can also be calculated along the height axis in the following way:

[0064] Where E'_h is the first sub-feature vector obtained by decomposing the second feature vector along the height axis.

[0065] Step a4: For any one of the plurality of second sub-feature vectors, calculate the second relevance score between the query vector and the second sub-feature vector; Step a5: Normalize the multiple second correlation scores corresponding to the multiple second sub-feature vectors to obtain multiple second probability values; the sum of the multiple second probability values ​​is 1; Step a6: The second attention result is obtained by weighted summation of the multiple second sub-feature vectors using the multiple second probability values.

[0066] In steps a4-a6 above, attention can also be calculated along the width axis in the following way:

[0067] Where E'_h is the second sub-feature vector obtained by decomposing the second feature vector along the width axis.

[0068] Step S106: Concatenate the query vector with the first attention result of the plurality of first sub-feature vectors and the second attention result of the query vector with the plurality of second sub-feature vectors to obtain the third feature vector.

[0069] Specifically, the query vector represents the channel feature parameter estimation task and is used to determine the feature information most relevant to channel parameter estimation from the second feature vector. The concatenated third feature vector can capture multi-dimensional channel feature dependencies such as path propagation in height and width.

[0070] In this embodiment, the channel query vector can be understood as a channel parameter estimation task, such as estimating path loss or angle of arrival. For example, when the query vector interacts with the second feature vector output by the encoder, the similarity of different feature information in the second feature vector is calculated. Then, the focus is on the feature information most relevant to the current estimation task. For example, when estimating AoA, more attention should be paid to the direction of the reflector; when estimating path loss, more attention should be paid to the transceiver distance and obstacle density. These similarities are then converted into attention weights to extract and integrate the most critical contextual information. In this way, the model can perform this information extraction and integration in different dimensions such as height and width to capture the nonlinear relationship between feature information and channel parameters.

[0071] In this embodiment of the application, the axial cross-attention mechanism in steps S105-S106 can learn and reveal the correspondence between physical features (i.e., feature information of multiple obstacles in the indoor environment) and channel parameters. For example, the location information of the obstacle can determine AoA (angle of arrival) and ToA (time of arrival); the material information of the obstacle can determine the path loss and reflection coefficient; the shape information of the obstacle can determine the AoD / AoA change and scattering intensity.

[0072] Step S107: Estimate key channel parameters using the third feature vector. The key channel parameters include path loss, arrival time, arrival angle, and departure angle.

[0073] In some specific embodiments, during the estimation of key channel parameters, partial feature information of obstacles with weights higher than a preset weight threshold in the third feature vector can be determined. Then, the key channel parameters can be estimated using this partial feature information of the obstacles. For example, path loss can be determined based on the reflection coefficient, which is determined by the material information of the obstacle. Therefore, path loss can be determined by the material information of the obstacle. Another example is that arrival time can be determined based on the shortest reflection path length, and arrival angle can be determined based on the incident / reflection direction of the dominant reflection point. The shortest reflection path length and the incident / reflection direction of the dominant reflection point are determined by the location information of the obstacle. Therefore, arrival time and arrival angle can be determined by the location information of the obstacle.

[0074] In this embodiment, sparse self-attention is mainly used to focus on key obstacles, as only key obstacles can affect signal propagation; the axial cross-attention mechanism can map the spatial structure (i.e., the feature information of multiple obstacles in the indoor environment) into the parameter space, where each channel parameter is determined by specific spatial features. The encoder-decoder structure is used to simulate the physical process of "propagation environment → channel response".

[0075] Corresponding to the above channel reconstruction methods, this application also provides a channel reconstruction model, which includes an encoder and a decoder; The encoder is used to acquire a first feature vector, which includes feature information of multiple obstacles in the indoor environment, transmitter position information, and receiver position information; the feature information includes shape information, position information, and material information; based on the hash value of the feature information of each obstacle in the first feature vector, the feature information of the multiple obstacles is assigned to multiple feature buckets; for multiple target obstacles in the same feature bucket or adjacent feature buckets, the attention result of the feature information of the multiple target obstacles is calculated; and all attention results corresponding to the multiple feature buckets are concatenated to obtain a second feature vector; The decoder is used to decompose the second feature vector along the height axis and the width axis to obtain multiple first sub-feature vectors and multiple second sub-feature vectors respectively; each first sub-feature vector is used to characterize the feature information of the multiple obstacles in the same feature dimension; each second sub-feature vector is used to characterize the feature information of the corresponding obstacle in multiple feature dimensions; the query vector is concatenated with the first attention result of the multiple first sub-feature vectors and the second attention result of the query vector with the multiple second sub-feature vectors to obtain a third feature vector; the query vector characterizes the channel parameter estimation task and is used to determine the feature information most relevant to the channel parameter estimation from the second feature vector; the third feature vector is used to estimate key channel parameters, including path loss, time of arrival, angle of arrival, and departure angle.

[0076] In some specific embodiments, the model structure of the channel reconstruction model is as follows: Figure 2 As shown.

[0077] In some specific embodiments, the channel reconstruction model stacks a 6-layer encoder-decoder architecture, adding positional coding to each layer:

[0078]

[0079] Where pos is the position and i is the dimension index. This refers to the model dimension. Add PE to the input: This preserves spatial order information. Finally, after processing by the Transformer, the channel feature vector is output. .

[0080] In some specific embodiments, the channel reconstruction model is trained through the following steps: Step b1: Construct the multi-task loss function.

[0081] The multi-task loss function is shown in the following formula:

[0082] in, The estimated loss representing path loss. Estimated loss indicating arrival time Estimated loss representing the angle of arrival Indicates the estimated loss from the departure angle; Loss weights representing path loss The loss weight representing the arrival time The loss weight representing the angle of arrival This represents the loss weight based on the departure angle.

[0083] In some specific embodiments, the weight λ is optimized through grid search. =0.5, =0.3, =0.2, using the AdamW optimizer, with an initial learning rate of 0.001.

[0084] Step b2: The channel reconstruction model is trained iteratively through the multi-task loss function until the difference between the key channel parameters output by the decoder and the target label converges to within a preset threshold.

[0085] Specifically, the preset threshold can be set according to the actual situation, and no specific limitation is made here.

[0086] In some specific embodiments, the target label includes a real label and a pseudo label; the pseudo label is generated based on environmental data and ray tracing parameters through a simulated signal propagation path.

[0087] Specifically, in an indoor environment, RF measurement equipment can be used to collect signal data at multiple receiving points, including path loss PL, arrival time ToA, arrival angle AoA, and departure angle AoD, and these data can be labeled with real tags. Alternatively, ray tracing simulation software (Remcom Wireless InSite) can be used to generate pseudo-tags: input the collected environmental data, set the ray tracing parameters (maximum number of reflections 3, frequency 28GHz), simulate the signal propagation path, and generate pseudo-tags corresponding to the real tags. To form the overall tag .

[0088] In some specific embodiments, the model training further includes: automatically reducing the learning rate during the training process by cosine annealing learning rate scheduling, so that the channel reconstruction model can obtain a better local minimum during the convergence phase, thereby improving the joint estimation accuracy of path loss, arrival time, arrival angle and departure angle.

[0089] Specifically, the cosine annealing learning rate scheduling is shown in the following formula:

[0090] Where t is the current iteration and T is the total number of iterations. Training continues until the loss converges (set to 100 epochs), and the final estimated channel data, including the values ​​of PL, ToA, AoA, and AoD, is output to complete the overall channel reconstruction.

[0091] This application proposes an indoor high-frequency channel reconstruction method based on an improved Transformer architecture. The core of this application is the design of an end-to-end deep learning framework that directly estimates key channel parameters, including path loss, time of arrival (ToA), angle of arrival (AoA), and angle of departure (AoD), by inputting the physical characteristics of the indoor environment (such as object locations, material properties, and spatial layout). This method avoids the computational overhead of traditional models.

[0092] This application employs a sparse self-attention mechanism, which significantly reduces computational complexity and improves efficiency in processing high-dimensional data by dynamically selecting a subset of relevant environmental features. The protected aspects include the specific implementation of the sparse attention algorithm and its application in environmental feature extraction to ensure the model's feasibility on resource-constrained devices.

[0093] This application introduces an axial cross-attention mechanism, which models the complex interaction between obstacle feature information and channel parameters along different dimensions (such as the height axis and the width axis), which can greatly improve the estimation accuracy of channel features.

[0094] The channel reconstruction method provided in this application is a channel reconstruction scheme based on an enhanced transformer. It reduces the quadratic complexity of attention computation through a sparse self-attention mechanism, from O(n...) 2The computational complexity is reduced to O(n log n). However, existing models and methods tend to overlook complex interactions, such as multipath reflections, when processing feature information of obstacles in high-dimensional indoor environments. Our proposed axial cross-attention mechanism better captures axial dependencies, improves the estimation accuracy of parameters such as AoA and AoD, and adapts to diverse indoor scenarios. Existing technologies, on the other hand, are sensitive to environmental changes and require recalibration.

[0095] Corresponding to the above implementation methods of channel reconstruction, this application also provides a channel reconstruction apparatus for executing the channel reconstruction methods described in the above embodiments. For example... Figure 3 As shown, the channel reconstruction device includes: The first feature vector construction module is used to construct a first feature vector, which includes feature information of multiple obstacles in the indoor environment, transmitter location information, and receiver location information; the feature information includes shape information, location information, and material information. The feature information allocation module is used to allocate the feature information of the multiple obstacles to multiple feature buckets according to the hash value of the feature information of each obstacle in the first feature vector; The attention result calculation module is used to calculate the attention result of the feature information of multiple target obstacles in the same feature bucket or adjacent feature buckets; The second feature vector generation module is used to concatenate all attention results corresponding to the multiple feature buckets to obtain the second feature vector; The second feature vector decomposition module is used to decompose the second feature vector along the height axis and the width axis to obtain multiple first sub-feature vectors and multiple second sub-feature vectors respectively; each first sub-feature vector is used to characterize the feature information of the multiple obstacles in the same feature dimension; each second sub-feature vector is used to characterize the feature information of the corresponding obstacle in multiple feature dimensions; The third feature vector generation module is used to concatenate the query vector with the first attention results of the plurality of first sub-feature vectors and the second attention results of the query vector with the plurality of second sub-feature vectors to obtain a third feature vector; the query vector represents the parameter estimation task and is used to determine the feature information most relevant to channel parameter estimation from the second feature vector; The critical channel parameter estimation module is used to pre-estimate critical channel parameters using the third feature vector. The critical channel parameters include path loss, time of arrival, angle of arrival, and departure angle.

[0096] Optionally, the first feature vector construction module is further configured to, for any one of the plurality of obstacles, acquire the shape information, position information, and material information of the obstacle; determine the transmitter position information and the receiver position information; use the shape information, the position information, the material information, the transmitter position information, and the receiver position information as feature information of the obstacle; and construct the first feature vector based on the feature information of the plurality of obstacles.

[0097] Optionally, the feature information allocation module is further configured to calculate the hash value of the feature information of each obstacle using a hash function; for any two obstacles among the plurality of obstacles, if the hash values ​​of the two obstacles are the same, the feature information of the two obstacles is allocated to the same feature bucket; if the difference between the hash values ​​of the two obstacles is a preset difference threshold, the feature information of the two obstacles is allocated to adjacent feature buckets.

[0098] Optionally, the device also includes: The first attention result calculation module is used to calculate a first relevance score between the query vector and the first sub-feature vector for any one of the plurality of first sub-feature vectors; normalize the plurality of first relevance scores corresponding to the plurality of first sub-feature vectors to obtain a plurality of first probability values; the sum of the plurality of first probability values ​​is 1; and perform a weighted summation of the plurality of first sub-feature vectors using the plurality of first probability values ​​to obtain the first attention result. The second attention result calculation module is used to calculate a second relevance score between the query vector and the second sub-feature vector for any one of the plurality of second sub-feature vectors; normalize the plurality of second relevance scores corresponding to the plurality of second sub-feature vectors to obtain a plurality of second probability values; the sum of the plurality of second probability values ​​is 1; and perform a weighted summation of the plurality of second sub-feature vectors using the plurality of second probability values ​​to obtain a second attention result.

[0099] The channel reconstruction apparatus and the channel reconstruction method provided in the above embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the applications stored therein.

[0100] This application also provides a computer device for performing the channel reconstruction method described above. Please refer to... Figure 4 This illustrates a schematic diagram of a computer device provided by some embodiments of this application. For example... Figure 4As shown, the computer device 4 includes: a processor 400, a memory 401, a bus 402, and a communication interface 403. The processor 400, the communication interface 403, and the memory 401 are connected via the bus 402. The memory 401 stores a computer program that can run on the processor 400. When the processor 400 runs the computer program, it executes the channel reconstruction method provided in any of the foregoing embodiments of this application.

[0101] The memory 401 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 403 (which can be wired or wireless), such as the Internet, wide area network, local area network, or metropolitan area network.

[0102] Bus 402 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. Memory 401 is used to store programs. After receiving an execution instruction, processor 400 executes the program. The channel reconstruction method disclosed in any of the foregoing embodiments can be applied to processor 400, or implemented by processor 400.

[0103] The processor 400 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 400 or by instructions in software form. The processor 400 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 401. The processor 400 reads the information in memory 401 and, in conjunction with its hardware, completes the steps of the above method.

[0104] The computer device provided in this application embodiment and the channel reconstruction method provided in this application embodiment are based on the same inventive concept and have the same beneficial effects as the methods they adopt, operate or implement.

[0105] This application also provides a computer-readable storage medium corresponding to the channel reconstruction method provided in the foregoing embodiments. Please refer to... Figure 5 The computer-readable storage medium shown is an optical disc 30, on which a computer program (i.e., a program product) is stored. When the computer program is run by a processor, it executes the channel reconstruction method provided in any of the foregoing embodiments.

[0106] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical and magnetic storage media, which will not be elaborated here.

[0107] The computer-readable storage medium provided in the above embodiments of this application and the channel reconstruction method provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the applications stored therein.

[0108] It should be noted that: Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of this application may be practiced without these specific details. In some instances, well-known structures and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0109] Similarly, it should be understood that, for the sake of brevity and to aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of this application, various features of this application are sometimes grouped together in a single embodiment, figure, or description thereof. However, this disclosure should not be construed as reflecting a schematic diagram in which the claimed application requires more features than expressly recited in each claim. Rather, as reflected in the following claims, inventive aspects lie in fewer than all features of a single foregoing disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into that detailed description, wherein each claim itself is a separate embodiment of this application.

[0110] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features but not others included in other embodiments, combinations of features from different embodiments are intended to be within the scope of this application and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination.

[0111] The above description is merely a preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A channel reconstruction method, characterized in that, The method includes: A first feature vector is constructed, which includes feature information of multiple obstacles in the indoor environment; the feature information includes shape information, position information, and material information. Based on the hash value of the feature information of each obstacle in the first feature vector, the feature information of the multiple obstacles is allocated to multiple feature buckets; For multiple target obstacles within the same feature bucket or adjacent feature buckets, calculate the attention results of the feature information of the multiple target obstacles; The second feature vector is obtained by concatenating all the attention results corresponding to the multiple feature buckets. The second feature vector is decomposed along the height axis and the width axis to obtain multiple first sub-feature vectors and multiple second sub-feature vectors; each first sub-feature vector is used to characterize the feature information of the multiple obstacles in the same feature dimension; each second sub-feature vector is used to characterize the feature information of the corresponding obstacle in multiple feature dimensions; The query vector is concatenated with the first attention result of the plurality of first sub-feature vectors and the second attention result of the query vector with the plurality of second sub-feature vectors to obtain a third feature vector; the query vector represents the parameter estimation task and is used to determine the feature information most relevant to channel parameter estimation from the second feature vector; The key channel parameters, including path loss, time of arrival, angle of arrival, and angle of departure, are estimated using the third feature vector.

2. The method according to claim 1, characterized in that, Constructing the first feature vector includes: For any one of the plurality of obstacles, obtain the shape information, position information, and material information of the obstacle; Determine the transmitter location information and receiver location information; The shape information, the position information, the material information, the transmitter position information, and the receiver position information are used as the feature information of the obstacle; The first feature vector is constructed based on the feature information of the multiple obstacles.

3. The method according to claim 1 or 2, characterized in that, Based on the hash value of the feature information of each obstacle in the first feature vector, the feature information of the multiple obstacles is allocated to multiple feature buckets, including: The hash value of the feature information of each obstacle is calculated using a hash function; For any two obstacles among the plurality of obstacles, if the hash values ​​of the two obstacles are the same, then the feature information of the two obstacles is assigned to the same feature bucket; If the difference between the hash values ​​of the two obstacles is equal to a preset difference threshold, then the feature information of the two obstacles will be assigned to adjacent feature buckets.

4. The method according to claim 1 or 2, characterized in that, After decomposing the second feature vector along the height axis and the width axis to obtain multiple first sub-feature vectors and multiple second sub-feature vectors respectively, the method further includes: For any one of the plurality of first sub-feature vectors, calculate the first relevance score between the query vector and the first sub-feature vector; The multiple first correlation scores corresponding to the multiple first sub-feature vectors are normalized to obtain multiple first probability values; the sum of the multiple first probability values ​​is 1; The first attention result is obtained by weighted summation of the multiple first sub-feature vectors using the multiple first probability values; For any one of the plurality of second sub-feature vectors, calculate a second relevance score between the query vector and the second sub-feature vector; The multiple second correlation scores corresponding to the multiple second sub-feature vectors are normalized to obtain multiple second probability values; the sum of the multiple second probability values ​​is 1; The second attention result is obtained by weighted summation of the multiple second sub-feature vectors using the multiple second probability values.

5. A channel reconstruction model, characterized in that, The channel reconstruction model includes: an encoder and a decoder; The encoder is used to acquire a first feature vector, which includes feature information of multiple obstacles in the indoor environment, transmitter position information, and receiver position information; the feature information includes shape information, position information, and material information; based on the hash value of the feature information of each obstacle in the first feature vector, the feature information of the multiple obstacles is assigned to multiple feature buckets; for multiple target obstacles in the same feature bucket or adjacent feature buckets, the attention result of the feature information of the multiple target obstacles is calculated; and all attention results corresponding to the multiple feature buckets are concatenated to obtain a second feature vector; The decoder is used to decompose the second feature vector along the height axis and the width axis to obtain multiple first sub-feature vectors and multiple second sub-feature vectors respectively; each first sub-feature vector is used to characterize the feature information of the multiple obstacles in the same feature dimension; each second sub-feature vector is used to characterize the feature information of the corresponding obstacle in multiple feature dimensions; the query vector is concatenated with the first attention result of the multiple first sub-feature vectors and the second attention result of the query vector with the multiple second sub-feature vectors to obtain a third feature vector; the query vector characterizes the channel parameter estimation task and is used to determine the feature information most relevant to channel parameter estimation from the second feature vector; the key channel parameters are found through the third feature vector, and the key channel parameters include path loss, time of arrival, angle of arrival, and departure angle.

6. The channel reconstruction model according to claim 5, characterized in that, The channel reconstruction model is trained through the following steps: Construct a multi-task loss function, which includes estimated losses for path loss, arrival time, arrival angle, and departure angle, as well as loss weights for each estimated loss. The channel reconstruction model is trained iteratively through the multi-task loss function until the difference between the key channel parameters output by the decoder and the target label converges to within a preset threshold. The target labels include real labels and pseudo labels; the pseudo labels are generated based on environmental data and ray tracing parameters through a simulated signal propagation path.

7. The channel reconstruction model according to claim 6, characterized in that, The model training also includes: By using cosine annealing to automatically reduce the learning rate during training, the channel reconstruction model can obtain a better local minimum during the convergence phase, thereby improving the joint estimation accuracy of path loss, arrival time, arrival angle, and departure angle.

8. A channel reconstruction apparatus, characterized in that, The device includes: The first feature vector construction module is used to construct a first feature vector, which includes feature information of multiple obstacles in the indoor environment, transmitter location information, and receiver location information; the feature information includes shape information, location information, and material information. The feature information allocation module is used to allocate the feature information of the multiple obstacles to multiple feature buckets according to the hash value of the feature information of each obstacle in the first feature vector; The attention result calculation module is used to calculate the attention result of the feature information of multiple target obstacles in the same feature bucket or adjacent feature buckets; The second feature vector generation module is used to concatenate all attention results corresponding to the multiple feature buckets to obtain the second feature vector; The second feature vector decomposition module is used to decompose the second feature vector along the height axis and the width axis to obtain multiple first sub-feature vectors and multiple second sub-feature vectors respectively; each first sub-feature vector is used to characterize the feature information of the multiple obstacles in the same feature dimension; each second sub-feature vector is used to characterize the feature information of the corresponding obstacle in multiple feature dimensions; The third feature vector generation module is used to concatenate the query vector with the first attention results of the plurality of first sub-feature vectors and the second attention results of the query vector with the plurality of second sub-feature vectors to obtain a third feature vector; the query vector represents the parameter estimation task and is used to determine the feature information most relevant to channel parameter estimation from the second feature vector; The critical channel parameter estimation module is used to estimate critical channel parameters through the third feature vector, wherein the critical channel parameters include path loss, time of arrival, angle of arrival, and departure angle.

9. A computer device, characterized in that, include: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the channel reconstruction method of any one of claims 1 to 4.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the channel reconstruction method according to any one of claims 1 to 4.

Images