A path loss prediction method considering Fresnel zone characteristics

Abstract

The present invention discloses a path loss prediction method considering the characteristics of the Fresnel zone, and belongs to the field of wireless communication technology. The method includes the following steps: generating image features: generating environmental image features through an urban building environment image feature generator; extracting features: extracting characteristic parameters related to path loss by training a feature extraction network; predicting path loss: generating a path loss prediction value. Compared with the prior art, the present invention has the following advantages: the method improves the traditional path loss prediction method based on environmental feature maps by considering the influence of obstacle distribution in the Fresnel zone on electromagnetic propagation characteristics, and enhances the model's ability to extract environmental features; the present invention can achieve accurate path loss prediction for urban building scenes, and can achieve high-precision prediction at higher carrier frequency bands, providing a new solution for the demand for high-precision channel modeling in future wireless communication network construction.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and more specifically, to a path loss prediction method considering Fresnel zone characteristics. Background Art

[0002] As the demand for ultra-high data rate transmission, ultra-low latency, global coverage, and smart services continues to increase, the existing 5G wireless communication networks are no longer sufficient to meet these demands. Driven by the demand for higher performance, 6G networks will need to leverage new enabling technologies to meet the challenges posed by emerging application scenarios, such as integrated air, space, land, and sea networks. This inevitably puts forward higher standards for the development and optimization of future wireless communication networks. High-precision wireless channel modeling is the basis for designing wireless communication systems, and the modeling of path loss is an indispensable part of it, playing an important role in the construction of future communication networks. Therefore, it is necessary to further improve the accuracy of path loss prediction to meet the growing demand for 6G network construction.

[0003] Channel modeling based on artificial intelligence (AI) technology is an important direction of 6G channel research. As a popular method in the field of AI, deep learning (DL) has a broad research space for its integration with channel modeling. Traditional modeling methods such as statistical modeling and deterministic modeling are usually difficult to obtain or extract effective environmental parameters. The DL-based channel modeling method can flexibly configure scene features, use machine vision to efficiently extract environmental features, and learn the laws of radio wave propagation from them, so as to achieve high-accuracy modeling and good model generalization ability.

[0004] However, how to achieve in-depth and effective environmental feature extraction is still an important issue for DL-based channel modeling. Considering that the distribution of obstacles in the Fresnel zone is an important factor affecting radio wave propagation, when the occlusion rate of the first Fresnel zone is less than 20%, the communication link can be regarded as a free space. At the same time, it is a feature suitable for visualization, which means that the environmental feature design for the Fresnel zone can provide prior knowledge of radio wave propagation for the DL-based path loss prediction model, while assisting it to more effectively extract environmental features, thereby improving the prediction accuracy of the model.

[0005] In the related technology, for example, Chinese patent CN116318462A provides a refined path loss prediction method for dual-modal network correction. This method uses a path loss empirical model to give statistical prediction values ​​and completes the common characterization of path loss. After that, a dual-modal network is used to learn the specific propagation characteristics of the communication environment from the visual and feature modes, obtain the correction value of the path loss empirical model, and complete the specific description of the path loss at different locations. However, this solution does not take into account that the distribution of obstacles in the Fresnel zone is an important factor affecting radio wave propagation.

[0006] From the above, it can be seen that the relevant technology does not provide any technical inspiration on how to achieve more effective environmental feature extraction and improve model prediction accuracy based on DL-based channel modeling and considering the environmental characteristics of the Fresnel zone. Summary of the invention

[0007] 1. Technical issues to be solved

[0008] In response to the problem in the prior art of how to achieve more effective environmental feature extraction and improve model prediction accuracy based on DL-based channel modeling and considering the environmental characteristics of the Fresnel zone, the present invention provides a path loss prediction method considering the Fresnel zone characteristics, which can realize accurate prediction of path loss in urban building scenes using AI methods.

[0009] 2. Technical solution

[0010] The purpose of the present invention is achieved through the following technical solutions.

[0011] A path loss prediction method considering the characteristics of the Fresnel zone comprises the following steps:

[0012] Generate image features: Generate environmental image features through the urban building environment image feature generator;

[0013] Extract features: Extract feature parameters related to path loss by training the feature extraction network;

[0014] Predicting path loss: Generate a path loss prediction value through the path loss predictor and the extracted characteristic parameters.

[0015] Furthermore, the step of generating image features is specifically as follows: using an urban building environment image feature generator and an urban building environment image feature generation algorithm, the environment image features are generated based on urban building distribution data and real channel parameters.

[0016] Furthermore, the urban building environment image feature generation algorithm includes: inputting the building distribution image and measured channel parameters of the target urban area into the urban building environment image feature generator;

[0017] The urban building environment image feature generator calculates the Fresnel zone size based on the building distribution image and the measured channel parameters, and generates the Fresnel zone mask image;

[0018] The urban building environment image feature generator calculates the free space path loss range of the target area according to the measured channel parameters and generates a free space path loss distribution image.

[0019] Furthermore, the step of generating the Fresnel zone mask image is: determining the range of the target area and the size of the generated feature image according to the building distribution image, and calculating the range of the Fresnel zone; the calculation formula of the Fresnel zone radius P is:

[0020]

[0021] Where d1 and d2 represent the distances between the receiver, transmitter and obstacles, and λ represents the carrier frequency.

[0022] The calculated Fresnel zone area range is converted into a corresponding pixel range, and a Fresnel zone mask image is drawn.

[0023] Furthermore, the steps of generating the free space path loss distribution image are: finding the maximum distance between the transceiver and the transmitter in the target data set, calculating the path loss corresponding to the distance as the maximum pixel path loss, and calculating the pixel values ​​corresponding to the path loss between different pixel points and the transmitter by interpolation;

[0024] The actual step length of the distance represented by a unit pixel is determined through the building distribution image. With the transmitter as the center, a free space path loss distribution map is generated layer by layer outward according to the pixel step length.

[0025] Furthermore, the formula for calculating the pixel value corresponding to the path loss between different pixel points and the transmitter is:

[0026]

[0027] Among them, C i Represents the pixel value of the target position pixel, P i represents the free space path loss value corresponding to the pixel at the target location, P min and P max They represent the minimum free space path loss of the dataset and the maximum free space path loss of the dataset, respectively.

[0028] Furthermore, the feature extraction step is specifically as follows: training a feature extraction network through channel features and generated environment image features to extract feature parameters related to path loss;

[0029] Among them, the extracted features include environmental image feature extraction and channel feature extraction.

[0030] Furthermore, the environmental image feature extraction includes: the environmental image feature extraction process receives the building distribution image and the environmental image features generated by the urban building environmental image feature generator to generate a two-dimensional environmental feature vector.

[0031] Furthermore, the channel feature extraction includes: receiving measured channel parameters and outputting a channel feature extraction vector.

[0032] Furthermore, the step of predicting the path loss is specifically: receiving the generated environment feature extraction vector and channel feature extraction vector;

[0033] Concatenate the environment feature extraction vector and the channel feature extraction vector;

[0034] The (multi-layer perceptron) MLP network architecture is used to fuse and extract the spliced ​​features and output the path loss prediction results.

[0035] 3. Beneficial effects

[0036] Compared with the prior art, the advantages of the present invention are:

[0037] (1) The path loss prediction method considering the Fresnel zone characteristics of the present invention can extract the environmental characteristics of the urban building scene, perform high-dimensional feature extraction on it, and learn the potential relationship between the environmental characteristics and the path loss, and finally realize the path loss prediction of the target location; the results obtained by this solution can improve the feature extraction and learning ability of the AI-based prediction model for the real environment, which is helpful for the design and performance evaluation of future communication systems;

[0038] (2) The present invention verifies and evaluates the prediction accuracy of the model by calculating the absolute error and root mean square error between the predicted path loss and the measured data;

[0039] (3) The present invention proposes a neural network prediction model based on CNN and MLP, which greatly reduces the complexity of path loss prediction for specific urban building environments. After the model is trained, it can perform fast and accurate path loss prediction based on externally provided building distribution data and simple channel parameter settings. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A schematic diagram of the main model components and their working principles in one embodiment of the present invention;

[0041] Figure 2 A schematic diagram of an input image feature sample for a feature extraction process according to an embodiment of the present invention;

[0042] Figure 3Another schematic diagram of an input image feature sample for a feature extraction process according to an embodiment of the present invention;

[0043] Figure 4 Another schematic diagram of an input image feature sample for a feature extraction process according to an embodiment of the present invention;

[0044] Figure 5 Schematic diagram of a network architecture of feature extraction and path loss predictor according to an embodiment of the present invention;

[0045] Figure 6 A schematic diagram of a satellite image of a measured environment and a measured route in one embodiment of the present invention;

[0046] Figure 7 It is a schematic diagram of the comparison between the path loss prediction results and the actual measurement results in different frequency bands in one embodiment of the present invention;

[0047] Figure 8 Another schematic diagram showing the comparison between the path loss prediction results and the actual measurement results in different frequency bands in one embodiment of the present invention;

[0048] Fig. 9 Another schematic diagram showing the comparison between the path loss prediction results and the actual measurement results in different frequency bands in one embodiment of the present invention;

[0049] Fig.10 A schematic diagram of the distribution of the absolute error of path loss prediction in the 3.5 GHz frequency band along the measurement route in one embodiment of the present invention;

[0050] Fig.11 Another schematic diagram of the distribution of the absolute error of path loss prediction in the 3.5 GHz frequency band along the measurement route in one embodiment of the present invention;

[0051] Fig.12 Schematic diagram of path loss prediction error probability distribution in one embodiment of the present invention. DETAILED DESCRIPTION

[0052] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0053] like Figure 1 As shown, the present invention comprises an urban building environment image feature generator and a feature extractor and path loss predictor. The urban building environment image feature generator uses the building feature image to generate an environment feature map through the feature generator, and the feature extractor and path loss predictor receives the pre-configured channel parameter features and the environment feature map, and uses the deep neural network MLP and CNN to obtain and splice the feature vectors respectively, and finally realizes the prediction of path loss through the MLP network.

[0054] Combination Figures 1 to 12, a path loss prediction method considering the Fresnel zone characteristics of the present invention comprises the following steps,

[0055] Generate image features:

[0056] Through the urban building environment image feature generator, using the urban building environment image feature generation algorithm, environmental image features are generated based on urban building distribution data and real channel parameters.

[0057] Specifically, combined with the Fresnel zone related principle, image features are generated that are conducive to model extraction and learning of environmental information. Among them, the urban building environment image feature generation algorithm includes the following steps:

[0058] (1) Inputting the building distribution image and measured channel parameters of the target urban area into the urban building environment image feature generator;

[0059] The building distribution image sample of the urban area is as follows: Figure 2 shown.

[0060] (2) The urban building environment image feature generator calculates the Fresnel zone size based on the building distribution image and the measured channel parameters, and generates a Fresnel zone mask image. The specific steps are as follows:

[0061] The scope of the target area and the size of the generated feature image are determined according to the building distribution image, and the scope of the Fresnel zone is calculated according to the calculation formula of the Fresnel zone radius.

[0062] Among them, the calculation formula of the Fresnel zone radius P is:

[0063]

[0064] Where d1 and d2 represent the distances between the receiver, transmitter and obstacles, and λ represents the carrier frequency.

[0065] Specifically, the calculated Fresnel zone area range is converted into a corresponding pixel range, and a Fresnel zone mask image is drawn.

[0066] Among them, the Fresnel zone mask image sample is as follows: Figure 3 shown.

[0067] (3) The urban building environment image feature generator calculates the free space path loss range of the target area based on the measured channel parameters and generates a free space path loss distribution image. The specific steps are as follows:

[0068] Find the maximum distance between the transceiver and the transmitter in the target data set, calculate the path loss corresponding to the distance as the maximum pixel path loss, and calculate the pixel values ​​corresponding to the path loss between different pixel points and the transmitter by interpolation. The specific calculation formula is:

[0069]

[0070] Among them, C i Represents the pixel value of the target position pixel, P i represents the free space path loss value corresponding to the pixel at the target location, P min and P max They represent the minimum free space path loss of the dataset and the maximum free space path loss of the dataset, respectively.

[0071] Specifically, the actual step length of the distance represented by a unit pixel is determined through the building distribution image, and a free space path loss distribution map is generated layer by layer outward according to the pixel step length with the transmitter as the center.

[0072] Among them, the free space path loss distribution pattern is as follows: Figure 4 shown.

[0073] Extract features:

[0074] The feature extraction network is trained by using the original features and generated features to extract feature parameters that are highly correlated with path loss.

[0075] Specifically, feature extraction includes environmental image feature extraction and channel feature extraction.

[0076] The environmental image feature extraction includes the following steps: the environmental image feature extraction process receives the building distribution image and the feature image generated by the urban building environment image feature generator to generate a two-dimensional environmental feature vector.

[0077] The environmental image feature extraction process used in this embodiment adopts a 6-layer CNN network structure. Combined with actual verification conditions, it can better realize the extraction of environmental feature vectors based on the building distribution feature map and the generated environmental feature image, while avoiding the decrease in generalization ability due to overfitting.

[0078] Specifically, the convolution kernel size of the first convolution layer is set to 5, and the size of each subsequent convolution kernel is set to 3. This setting helps to integrate features across multiple input channels, and can fuse the line of sight information provided by the Fresnel zone mask image and the distance data inherent in the free space path loss distribution image into the building environment features.

[0079] Specifically, the maximum pooling layer is set after the second, third, fifth, and sixth convolutional layers. This progressive adjustment aims to extract high-dimensional abstract features with discriminative and generalization capabilities.

[0080] Specifically, the ReLU (Rectified Linear Unit) function is selected as the activation function of the environmental image feature extraction process. The ReLU function is a commonly used activation function in artificial neural networks, which can alleviate the problem of gradient disappearance. The expression f(x) of the ReLU function is as follows:

[0081] f(x)=max(0,x) (3)

[0082] Wherein, x is the input of the ReLU function.

[0083] Specifically, the two-dimensional features are reshaped into a one-dimensional feature vector so as to be concatenated with the channel feature vector, and an environmental feature extraction vector is output.

[0084] In this embodiment, an adaptive full-play average pooling method is selected to replace the commonly used fully connected layer for feature reshaping, which greatly reduces the complexity and computational complexity of the model.

[0085] Specifically, the length of the output environment image feature extraction vector is 96 bits.

[0086] The channel feature extraction includes the following steps: receiving measured channel parameters and outputting a channel feature extraction vector.

[0087] The channel feature extraction process adopts a shuttle MLP network structure, which first expands the new features of the channel features and then compresses them, so that the network can learn a more complex nonlinear relationship between the original channel features and the target features, thereby enhancing the network's expression ability.

[0088] In this embodiment, ReLU is selected as the activation function of the channel feature extraction process.

[0089] Specifically, the length of the output channel feature extraction vector is 64 bits. This choice ensures that while environmental features play a dominant role in the prediction process, channel features will not be ignored because their proportion is too low.

[0090] Predicted path loss:

[0091] The path loss predictor generates a path loss prediction value using the extracted characteristic parameters.

[0092] The specific steps for predicting path loss are as follows:

[0093] (1) receiving an environment feature extraction vector and a channel feature extraction vector generated by a feature extractor;

[0094] (2) concatenating the environmental feature extraction vector and the channel feature extraction vector;

[0095] (3) Use the MLP (multi-layer perceptron) network architecture to fuse and extract the spliced ​​features and output the path loss prediction results.

[0096] Specifically, an inverted triangle MLP network structure is selected to quickly compress the concatenated features to improve the fitting efficiency while avoiding overfitting that affects the model accuracy and generalization ability.

[0097] Specifically, ReLU is selected as the activation function for the channel feature extraction process.

[0098] Specifically, the feature extraction and path loss predictor proposed in the present invention uses the Adam optimizer to update all parameters in the network, and the learning rate is set to 0.0012%.

[0099] The present invention compares the prediction results with the real channel data and the traditional AI prediction method, and proves the prediction accuracy based on the absolute error and its distribution and the root mean square error. In order to facilitate accurate evaluation with the real channel, all models are trained and verified using the same data, and the measured samples of the path loss prediction have the same longitude and latitude coordinates as the prediction results.

[0100] like Figure 5 As shown, the network architecture of the feature extraction and path loss predictor is shown. The feature extraction and path loss predictor in this scheme realizes the process of extracting features and predicting path loss.

[0101] The training model data used in this example are all collected by field measurements of the detection module. The detection module uses a combination of software and hardware, supplemented by calibration verification, and finally outputs the measured real channel data.

[0102] like Figure 6 As shown in Figure 1, satellite images of the actual measurement environment and the measured route are displayed. Figures 7 to 9 As shown in the figure, the path loss prediction results of different frequency bands are compared with the actual measurement results under the snapshot dimension. Figures 7 to 9 The predicted results at 0.69 GHz, 1.8 GHz and 3.5 GHz are compared with the path loss of the measured channel. The path loss of the real channel is shown as the blue curve in the figure, the predicted path loss of the present invention is shown as the red curve, and the path loss predicted by the traditional AI method is shown as the green curve.

[0103] By comparison, it can be seen that this scheme can accurately predict the changing trend of path loss in urban building environments. In the relatively low-frequency 0.69GHz and 1.8GHz bands, the prediction of the middle snapshot has a certain degree of absolute error due to severe building obstruction, with a root mean square error of 5.12dB, but in the relatively high frequency 3.5GHz, the prediction results of this scheme are highly consistent with the measured channel, with a root mean square error of 1.97dB. This is because the use of the Neel zone features directs the model's attention to radio propagation in this specific area, and at higher frequency bands, the building obstruction in the Neel zone is reduced, making the environmental features extracted by this scheme simpler and more effective. Therefore, this model of this scheme is easier to make accurate predictions in higher frequency bands.

[0104] like Figure 10 to Figure 11 As shown in the figure, the absolute error distribution of the model's predicted path loss and the measured channel data on the measurement route is shown in the 3.5 GHz frequency band. Fig.10 As shown in FIG. , the prediction error distribution of the model described in the present invention is as follows: Fig.11 As shown in , this is the prediction error distribution of the traditional AI method. Fig.12 The figure shows the distribution of the absolute error between the model's predicted path loss and the measured channel data. Figures 10 to 12 It can be seen more intuitively that in the urban building scenario, the prediction method proposed in this scheme can achieve higher prediction accuracy in line-of-sight sections and non-line-of-sight sections, and the prediction error distribution converges around 1dB with a higher probability.

[0105] The present invention combines the Fresnel zone theory to propose a path loss prediction method for urban building scenarios. In this method, by combining the AI ​​feature extraction method with the prior knowledge of electromagnetic propagation, the urban environment characteristics and channel characteristics are deeply extracted and used to predict the path loss. On this basis, combined with the measured data on campus, the proposed method is compared with the traditional AI prediction method, and the effectiveness and accuracy of the method are verified. The present invention can realize accurate path loss prediction for urban building scenarios, especially in higher carrier frequency bands, and can achieve high-precision prediction, which provides a new solution for the demand for high-precision channel modeling in the construction of future wireless communication networks.

[0106] The above schematically describes the invention and its implementation methods, which is not restrictive. Without departing from the spirit or basic features of the invention, the invention can be implemented in other specific forms. What is shown in the accompanying drawings is only one of the implementation methods of the invention. The actual structure is not limited thereto, and any figure mark in the claims should not limit the claims involved. Therefore, if a person of ordinary skill in the art is inspired by it, without departing from the purpose of the invention, a structural method and an embodiment similar to the technical solution are designed without creativity, which should all fall within the scope of protection of this patent. In addition, the word "including" does not exclude other elements or steps, and the word "one" before an element does not exclude the inclusion of "multiple" elements. The multiple elements stated in the product claim can also be implemented by one element through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any specific order.

Claims

1. A path loss prediction method considering Fresnel zone characteristics, comprising the following steps: Generate image features: Generate environmental image features based on urban building distribution data and real channel parameters through the urban building environment image feature generator and the urban building environment image feature generation algorithm; Feature extraction: Extract characteristic parameters related to path loss by training channel features and generating environmental image feature extraction networks; The feature extraction includes environmental image feature extraction and channel feature extraction. The environmental image feature extraction includes receiving the building distribution image and the environmental image features generated by the urban building environment image feature generator to generate a two-dimensional environmental feature extraction vector. The channel feature extraction includes receiving the measured channel parameters and outputting the channel feature extraction vector. Predicting path loss: receiving the generated environment feature extraction vector and channel feature extraction vector; concatenating the environment feature extraction vector and the channel feature extraction vector; fusing and extracting the concatenated features using a multi-layer perceptron (MLP) network architecture, and outputting the path loss prediction result; Among them, the urban building environment image feature generation algorithm includes: inputting the building distribution image and measured channel parameters of the target urban area into the urban building environment image feature generator; the urban building environment image feature generator calculates the Fresnel zone size according to the building distribution image and the measured channel parameters, and generates a Fresnel zone mask image; the urban building environment image feature generator calculates the free space path loss range of the target area according to the measured channel parameters, and generates a free space path loss distribution image.

2. The path loss prediction method according to claim 1, characterized in that: The steps of generating the Fresnel zone mask image are as follows: determining the range of the target area and the size of the generated feature image according to the building distribution image, and calculating the range of the Fresnel zone; the calculation formula of the Fresnel zone radius P is: Where d1 and d2 represent the distance between the receiver and the transmitter and the obstacle, and λ represents the carrier frequency; The calculated Fresnel zone area range is converted into a corresponding pixel range, and a Fresnel zone mask image is drawn.

3. The path loss prediction method according to claim 1, characterized in that: The steps of generating the free space path loss distribution image are as follows: finding the maximum distance between the transceiver and the transmitter in the target data set, calculating the path loss corresponding to the distance as the maximum pixel path loss, and calculating the pixel values ​​corresponding to the path loss between different pixel points and the transmitter by interpolation; The actual step length of the distance represented by a unit pixel is determined through the building distribution image. With the transmitter as the center, a free space path loss distribution map is generated layer by layer outward according to the pixel step length.

4. The path loss prediction method according to claim 3, characterized in that: The formula for calculating the pixel value corresponding to the path loss between different pixel points and the transmitter is: Among them, C i Represents the pixel value of the target position pixel, P i represents the free space path loss value corresponding to the pixel at the target location, P min and P max represent the minimum free space path loss and the maximum free space path loss of the dataset, respectively.

Images