A Method for Constructing a Knowledge Map Based on Integrating Environmental Topology Models

Abstract

This invention provides a method for constructing a channel knowledge map by integrating an environmental topology model, relating to the field of wireless communication technology. The method includes: Step S1, constructing environmental information conditions, including a signal strength sampling map, KNN interpolation enhancement input, a Loss map, a building height map within the task area, and a base station location map; Step S2, using the environmental information conditions as input, extracting deep environmental information through a convolutional network encoder and representing it as high-dimensional environmental topology information; Step S3, obtaining the environmental information vector corresponding to the target location based on the high-dimensional environmental topology information; Step S4, concatenating the target location and its corresponding environmental information vector as input, and using the received signal strength as a label, training a regression network; Step S5, predicting the received signal strength within the task area based on the trained regression network to obtain a channel knowledge map. This invention can effectively improve the construction accuracy of channel knowledge maps.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a method for constructing a WeChat knowledge map that integrates an environmental topology model. Background Technology

[0002] With the rapid development of low-altitude economy and vehicle-to-everything (V2X) applications, mobile communication networks are placing higher demands on efficient and accurate mobility management. In traditional downlink communication, received signal strength typically relies on terminal feedback, which suffers from high pilot overhead and information staleness due to feedback delay. Channel knowledge maps, by establishing a mapping relationship between terminal location and received signal strength, form a location-related channel knowledge database, which is expected to significantly reduce or even avoid the overhead of feedback-based received signal strength acquisition mechanisms, thereby addressing the aforementioned challenges.

[0003] Existing methods for constructing channel knowledge maps mainly fall into two categories: spatial interpolation-based methods and deep learning-based methods. Spatial interpolation-based methods (such as Kriging and K-nearest neighbors) heavily rely on dense data sampling, and their accuracy is often difficult to guarantee in complex wireless propagation environments. Deep learning-based methods, such as RadioUNet and RadioDiff, can fuse base station location, environmental topology models, and received signal strength sampling data to achieve more accurate channel knowledge map construction. However, these methods typically discretize environmental information into feature maps as input, resulting in the generated channel knowledge map being non-differentiable with respect to continuous location coordinates. This non-differentiability limits the applicability of channel knowledge maps in application scenarios requiring gradient-based optimization, such as trajectory design. On the other hand, although using regression networks such as multilayer perceptrons (MLPs) to directly regress received signal strength values ​​from continuous coordinates can naturally guarantee the differentiability of the channel knowledge map's location, these methods often struggle to effectively fuse complex environmental topology information, leading to less than ideal prediction accuracy in complex environments. Summary of the Invention

[0004] The main objective of this invention is to propose a method for constructing a channel knowledge map that integrates an environmental topology model, which can effectively improve the construction accuracy of the channel knowledge map.

[0005] This invention is achieved through the following technical solution: The method for constructing a WeChat knowledge map that integrates environmental topology models includes the following steps: Step S1: Construct environmental information conditions, which include a signal strength sampling map constructed by sampling the received signal strength of the base station at different locations in the task area, a KNN interpolation enhancement input constructed by performing KNN interpolation based on the sampling results, a LoS map constructed based on whether there is a direct link, a building height map and a base station location map in the task area. Step S2: Using environmental information conditions as input, extract deep environmental information through a convolutional network encoder and represent it as high-dimensional environmental topology information; Step S3: Obtain the environmental information vector corresponding to the target location based on the high-dimensional environmental topology information; Step S4: Concatenate the target location and its corresponding environmental information vector as input, and use the received signal strength at the target location as the label to train the regression network; Step S5: Based on the trained regression network, predict the received signal strength within the task area to obtain a WeChat knowledge map.

[0006] Furthermore, in step S1, the task area For one A rectangular urban area and They represent in x shaft and y The maximum index value of the center of the task region grid in the axial direction, at the sampling rate. In the mission area Different positions Collect the signal strength received by the base station To construct a signal strength sampling map ,in, This is the set of locations for sampling the received signal strength at the base station. .

[0007] Furthermore, in step S1, during KNN interpolation based on the sampling results, the total number of interpolation points is... In the i The interpolation result for each interpolation point is represented as follows: The KNN interpolation augmentation input is obtained from the interpolation results corresponding to all interpolation points. ,in, For the first i Interpolation target position corresponding to each interpolation point The K nearest neighbors, interpolation target position Its first k nearest neighbor points Euclidean distance.

[0008] Furthermore, in step S1, regarding the position... If the location has a direct link to the base station, then ,otherwise To construct the Loss map. .

[0009] Furthermore, in step S1, a building height map B is obtained based on an open-source map dataset, and a base station location map B is set according to the actual environment. BS The building height in map B is the first n Location The corresponding value is the building height at that location, as shown in Base Station Location Map B. BS Corresponding base station location The value at position 1 is set to 1, and the value at other positions is set to 0.

[0010] Furthermore, in step S2, the convolutional network encoder model structure is UNet, which contains 7 convolutional layers. The first 3 convolutional layers are downsampling layers, the 4th convolutional layer is an intermediate layer, and the last 3 convolutional layers are upsampling layers. Each convolutional layer uses ReLU as the activation function, and the number of channels output by the last convolutional layer is... The obtained high-dimensional environmental topology information is represented as ,in For dimension The real number space.

[0011] Furthermore, in step S3, feature values ​​corresponding to the target position q are extracted from each channel of the high-dimensional environmental topology information, and the feature values ​​corresponding to all channels are integrated to obtain the environmental information vector. .

[0012] Furthermore, in step S4, the regression network includes a multilayer perceptron or a KAN network.

[0013] Furthermore, in step S5, by freezing f condition Chinese f q The gradient is obtained by backpropagating the gradient of the base station received signal strength at the target location q. All locations in the task area are taken as target locations and steps S3 to S5 are performed to obtain the knowledge map.

[0014] As can be seen from the above description of the present invention, compared with the prior art, the present invention has the following beneficial effects: This invention uses a signal strength sampling map, KNN interpolation enhancement input, Loss map, building height map, and base station location map as environmental information conditions. These environmental information conditions are then used as input to a convolutional network encoder to obtain high-dimensional environmental topology information. Based on this high-dimensional topology information, an environmental information vector corresponding to the target location is extracted. The target location and its corresponding environmental information vector are concatenated and used as input. The received signal strength at the target location is used as a label to train a regression network. Finally, the trained regression network is used to predict the received signal strength within the task area, resulting in a channel knowledge map. This effectively integrates environmental information. Compared to traditional received signal strength regression methods that do not include environmental features, the channel knowledge map constructed in this invention exhibits lower normalized mean square error (NMSE) under various measurement sampling densities, effectively improving the construction accuracy of the channel knowledge map. Attached Figure Description

[0015] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] Figure 1 This is a flowchart of the present invention.

[0017] Figure 2 This is a comparison chart of NMSE at different sampling rates for the channel knowledge map constructed by the present invention and the comparison method.

[0018] Figure 3 This is a comparison diagram showing the effects of constructing a channel knowledge map using the present invention and a comparative method.

[0019] Figure 4 This is a schematic diagram of the gradient magnitude and phase of the channel knowledge map constructed in this invention (when the regression network uses MLP).

[0020] Figure 5 This is a schematic diagram of the gradient magnitude and phase of the channel knowledge map constructed in this invention (when the regression network uses KAN). Detailed Implementation

[0021] The present invention will be further described below through specific embodiments.

[0022] like Figure 1 As shown, the method for constructing a WeChat knowledge map by integrating an environmental topology model includes the following steps: Step S1: Construct environmental information conditions, which include a signal strength sampling map constructed by sampling the received signal strength of the base station at different locations in the task area, a KNN interpolation enhancement input constructed by performing KNN interpolation based on the sampling results, a LoS map constructed based on whether there is a direct link, a building height map and a base station location map in the task area. Specifically, the task area is represented as It is a A rectangular urban area and They represent in x shaft and y The maximum index value of the center of the task region grid in the axial direction, at the sampling rate. In the mission area Different positions Collect the signal strength received by the base station To construct a signal strength sampling map ,in, , The set of locations for sampling the received signal strength of the base station contains the following number of locations: For signal strength sampling map Its x-coordinate corresponds to The received signal strength is , The received signal strength is 0.

[0023] In this embodiment, X max =256, Y max =256, .

[0024] In the KNN interpolation based on the sampling results, the total number of interpolation points is: In the i The interpolation result for each interpolation point is represented as follows: The KNN interpolation augmentation input is obtained from the interpolation results corresponding to all interpolation points. ,in, For the first i Interpolation target position corresponding to each interpolation point The K nearest neighbors, interpolation target position Its first k nearest neighbor points The Euclidean distance. In this embodiment, K =4, For dimension The real number space.

[0025] For location If the location has a direct link to the base station, then ,otherwise To construct the Loss map. .

[0026] Obtain building height maps from open-source map datasets. Set up a base station location map according to the actual environment. The building height in map B is the first n Location The corresponding value is the building height at that location, as shown in Base Station Location Map B. BS Corresponding base station location The value at position 1 is set to 1, and the value at other positions is set to 0.

[0027] Integrating the above , B, B BS And L, constructing dimensions as Environmental information conditions.

[0028] Step S2: Using environmental information conditions as input, extract deep environmental information through a convolutional network encoder and represent it as high-dimensional environmental topology information; The convolutional network encoder model architecture is UNet, which contains 7 convolutional layers. The first 3 convolutional layers are downsampling layers, the 4th convolutional layer is an intermediate layer, and the last 3 convolutional layers are upsampling layers. Each convolutional layer uses ReLU as the activation function, and the output of the last convolutional layer has 10 channels. D F =128, the obtained high-dimensional environment topology information is represented as: ,in For dimension The real number space.

[0029] Step S3: Obtain the environmental information vector corresponding to the target location based on the high-dimensional environmental topology information; Specifically, feature values ​​corresponding to the target position q are extracted from each channel of the high-dimensional environmental topology information. By integrating the feature values ​​corresponding to all channels, an environmental information vector is obtained. .

[0030] Step S4: Concatenate the target location and its corresponding environmental information vector as input. The regression network is trained using the received signal strength at the target location as a label. The regression network includes a Multilayer Perceptron (MLP) or a KAN (Kolmogorov-Arnold) network. The regression network has an input layer dimension of 130 and an output layer dimension of 1. The MLP has hidden layer dimensions of 256, 256, 128, and the KAN has hidden layer dimensions of 10, 1. The B-spline function grid resolution is 8, and the order is 4. The specific process of training the regression network is based on existing techniques.

[0031] Step S5: Based on the trained regression network, predict the received signal strength within the task area to obtain a WeChat knowledge map. More specifically, by freezing f condition Chinese f q The gradient is obtained by backpropagating the gradient of the base station received signal strength at the target location q. All locations in the task area are taken as target locations and steps S3 to S5 are performed to obtain the knowledge map.

[0032] Figure 2 In the middle, the horizontal axis represents the sampling rate. The vertical axis represents NMSE. In the figure, "Environmental Topology Fusion-MLP Regression" refers to the invention with MLP regression network and "Environmental Topology Fusion-MLP Regression" refers to the invention with KAN regression network. It can be seen that, compared with other methods, the knowledge map constructed by the present invention has the smallest NMSE, that is, the highest accuracy.

[0033] Figure 3 The horizontal axis represents the target position. x Coordinates, with the vertical axis representing the target position. y As can be seen from the coordinates, compared with other methods, the knowledge map constructed by this invention can better capture environmental details and reconstruct the Loss of Space (LoS) region more accurately.

[0034] Figure 4 and Figure 5 The horizontal axis represents the target position. x Coordinates, with the vertical axis representing the target position. y As can be seen from the coordinates, this invention can achieve accurate gradient solving for position coordinates.

[0035] In this invention, the terms "first," "second," and "third," etc., are used only to distinguish similar objects and are not necessarily used to describe a specific order or sequence, nor should they be construed as indicating or implying relative importance. The use of terms such as "upper," "lower," "left," "right," "front," and "rear" to indicate orientation or positional relationships is based on the orientation or positional relationships shown in the accompanying drawings and is only for the convenience of describing the invention, not to indicate or imply that the device referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation on the scope of protection of this invention. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0036] Furthermore, in the description of this application, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0037] The above are merely specific embodiments of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial modifications made to the present invention using this concept shall be considered as infringing upon the protection scope of the present invention.

Claims

1. A method for constructing a WeChat knowledge map that integrates environmental topology models, characterized by: Includes the following steps: Step S1: Construct environmental information conditions, which include a signal strength sampling map constructed by sampling the received signal strength of the base station at different locations in the task area, a KNN interpolation enhancement input constructed by performing KNN interpolation based on the sampling results, a LoS map constructed based on whether there is a direct link, a building height map and a base station location map in the task area. Step S2: Using environmental information conditions as input, extract deep environmental information through a convolutional network encoder and represent it as high-dimensional environmental topology information; Step S3: Obtain the environmental information vector corresponding to the target location based on the high-dimensional environmental topology information; Step S4: Concatenate the target location and its corresponding environmental information vector as input, and use the received signal strength at the target location as the label to train the regression network; Step S5: Based on the trained regression network, predict the received signal strength within the task area to obtain a WeChat knowledge map. In step S5, freezing is performed. middle The gradient of the target position Perform gradient backpropagation to obtain the base station received signal strength for the target location. The gradient is used to treat all locations in the task area as target locations and proceed to steps S3 to S5 to obtain a WeChat knowledge map.

2. The method for constructing a WeChat knowledge map based on a fusion environment topology model according to claim 1, characterized in that: In step S1, the task area For one A rectangular urban area and These represent the maximum index values ​​at the center of the task region grid in the x-axis and y-axis directions, respectively, expressed as a sampling rate. In the mission area Different positions Collect the signal strength received by the base station To construct a signal strength sampling map ,in, This is the set of locations for sampling the received signal strength at the base station. .

3. The method for constructing a WeChat knowledge map based on a fusion environment topology model according to claim 2, characterized in that: In step S1, during KNN interpolation based on the sampling results, the total number of interpolation points is: The interpolation result at the i-th interpolation point is expressed as The KNN interpolation augmentation input is obtained from the interpolation results corresponding to all interpolation points. ,in, The interpolation target position corresponding to the i-th interpolation point The K nearest neighbors, interpolation target position Its k-th nearest neighbor Euclidean distance.

4. The method for constructing a WeChat knowledge map based on a fusion environment topology model according to claim 3, characterized in that: In step S1, for position If the location has a direct link to the base station, then ,otherwise To construct the Loss map. .

5. The method for constructing a WeChat knowledge map based on a fusion environment topology model according to claim 4, characterized in that: In step S1, a building height map is obtained based on an open-source map dataset. Set up a base station location map according to the actual environment. Building height map The nth position The corresponding value is the building height at that location; base station location map. Corresponding base station location The value at position 1 is set to 1, and the value at other positions is set to 0.

6. The method for constructing a WeChat knowledge map based on a fusion environment topology model according to any one of claims 2 to 5, characterized in that: In step S2, the convolutional network encoder model structure is UNet, which contains 7 convolutional layers. The first 3 convolutional layers are downsampling layers, the 4th convolutional layer is an intermediate layer, and the last 3 convolutional layers are upsampling layers. Each convolutional layer uses ReLU as the activation function, and the number of channels output by the last convolutional layer is... The obtained high-dimensional environmental topology information is represented as ,in For dimension The real number space.

7. The method for constructing a WeChat knowledge map based on a fusion environment topology model according to claim 6, characterized in that: In step S3, the target location is extracted from each channel of the high-dimensional environmental topology information. The feature values ​​of all channels are integrated to obtain the environmental information vector. .

8. The method for constructing a WeChat knowledge map based on a fusion environment topology model according to any one of claims 1 to 5, characterized in that: In step S4, the regression network includes a multilayer perceptron or a KAN network.

Images