Radio map collaborative estimation method based on generative adversarial network

Abstract

The invention discloses a radio map collaborative estimation method based on a generative adversarial network. The method comprises the following steps: S1, constructing a training data set; s2, adopting a UNet model as a generator, inputting a graphical geographical environment map and RSS information, and outputting a graphical radio map estimation result; s3, adopting a convolutional neural network model, namely a CNN model, as a discriminator for evaluating the authenticity of the radio map output by the generator; s4, in the model training stage, the generator and the discriminator are jointly optimized, that is, the generator and the discriminator are mutually optimized through an adversarial training process, the generator generates a radio map with higher quality, and the discriminator improves the discrimination capability of a generated result; and S5, in a reasoning stage, performing radio map estimation by using a generator. According to the invention, radio map estimation can be carried out based on received signal strength (RSS) information of a plurality of mobile users and a geographical environment map, and signal source information is not needed.

Description

Technical Field

[0001] The present invention relates to radio map estimation, and particularly to a collaborative radio map estimation method based on a generative adversarial network. Background Art

[0002] In the 6G era, real-time monitoring and management of radio resources are crucial for supporting diverse wireless applications. Traditional radio map estimation (RME) methods, such as data-driven interpolation methods and model-based ray tracing methods, cannot achieve both high accuracy and high speed simultaneously, and cannot realize accurate and fast radio map estimation. Existing deep learning-based methods are expected to solve the above problems, that is, by training a deep neural network with a high-precision dataset to enable the deep neural network to achieve fast and accurate radio map estimation.

[0003] However, current deep learning-based methods for achieving accurate radio map estimation all rely on signal source information, that is, signal source information is an indispensable part of the input information of the neural network. Since in a wireless network, the signal source may be a base station or various mobile devices, and its information (especially real-time information) is often unavailable. In addition, such estimation methods relying on signal source information have poor generalization ability when the geographical environment information is inaccurate. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a collaborative radio map estimation method based on a generative adversarial network, which can perform radio map estimation based on the received signal strength (RSS) information of multiple mobile users and the geographical environment map without signal source information.

[0005] The purpose of the present invention is achieved by the following technical solutions: A collaborative radio map estimation method based on a generative adversarial network, comprising the following steps: S1. Using the received signal strength information, RSS information, and geographical environment information of multiple users as data features, and using the radio map as a label to construct a training dataset, where the RSS information is the location information; S2. Using a UNet model as a generator, with the input being the graphical geographical environment map and RSS information, and the output being the graphical radio map estimation result; S3. Using a convolutional neural network model, that is, a CNN model as a discriminator to evaluate the authenticity of the radio map output by the generator; S4. In the model training stage, jointly optimizing the generator and the discriminator; S5. In the inference stage, using the generator for radio map estimation.

[0006] The beneficial effects of the present invention are as follows: The proposed collaborative radio map estimation (CRME) method based on generative adversarial network (GAN), namely GAN-CRME, can estimate the radio map based on the received signal strength (RSS) information of multiple mobile users and the geographical environment map without the need for signal source information. Based on the GAN learning algorithm, the accuracy and generalization ability of the system are improved. This method can work effectively even when the geographical data is inaccurate, automatically corrects some errors, and ensures the accuracy and robustness of the estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 is the flowchart of the method of the present invention; Figure 2 is a schematic diagram for estimating the radio map using the received signal strength, location of the user, and geographical environment map information; Figure 3 is a schematic diagram of the model training process based on the GAN framework; Figure 4 is a schematic diagram of the estimation accuracy of the present invention when there is no error in the physical environment map; Figure 5 is the estimation accuracy of the present invention when there is an error in the physical environment map (some buildings are missing, the blackened part). DETAILED DESCRIPTION OF THE EMBODIMENTS

[0008] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the protection scope of the present invention is not limited to the following.

[0009] This technical solution uses a dataset constructed based on the ray tracing method for deep neural network training. The differences from the existing work are as follows. First, we use the UNet network as the deep neural network model for radio map perception. Its feature is a graph-to-graph neural network that can deeply understand the physical environment and extract its features for image fusion generation. Second, we use the received signal strength information at the user end instead of the signal source information, which is a more practical solution. Finally, we design a training method based on generative adversarial network (GAN), use UNet as the generator, and introduce another CNN neural network model as the discriminator, and perform collaborative adversarial training on the two networks, greatly strengthening the performance of the two networks. The UNet network with enhanced performance can achieve accurate radio map estimation, and the gap between its accuracy and the accuracy of the data in the training set is very small, and it can be considered close to the accuracy of the ray tracing method. Specifically: As Figure 1 shown, a collaborative radio map estimation method based on generative adversarial network includes the following steps: S1. Take the received signal strength information, RSS information, and geographical environment information of multiple users as data features, and use the radio map as a label to construct a training dataset, where the RSS information is the location information; Use the received signal strength of the user, the location, and the geographical environment map information to estimate the idea framework of the radio map, as Figure 2 shown; Suppose the training dataset contains L data samples. For the th data sample: The data feature is , where represents the matrix composed of the received signal strength information and RSS information of all users, is the matrix representing the geographical environment information, and the matrix represents the corresponding radio map, that is, the label of the th data sample, where .

[0010] In the embodiment of the present application, the matrix is obtained by ray tracing method or manual measurement calculation. Assuming a region of 256m * 256m is considered and the radio map resolution is 1m * 1m, the dimensions of the above three matrices are all 256 * 256.

[0011] S2. Use the UNet model as the generator, with the input being the graphical geographical environment map and RSS information, and the output being the graphical radio map estimation result; S3. Use a convolutional neural network model, that is, the CNN model as the discriminator to evaluate the authenticity of the radio map output by the generator; For the specific models of the UNet generator and the CNN discriminator, select the 17 - layer UNet network as the generator, denoted as , and its parameter matrix is ; select the 5 - layer CNN network as the discriminator, denoted as , and its parameter matrix is .

[0012] S4. In the model training stage, the generator and the discriminator are jointly optimized, that is, through the adversarial training process, the generator and the discriminator optimize each other, the generator generates a higher - quality radio map, and the discriminator improves its discrimination ability for the generation result; The schematic diagram of the model training process based on the GAN framework is as Figure 3 shown: S401. For any data sample , input the data feature into the UNet generator to calculate the "false label" 。

[0013] S402. Pair the data features with the true labels as and input them into the CNN discriminator to calculate the scores 。

[0014] S403. Pair the data features with the "false labels" as and input them into the CNN discriminator to calculate the scores 。

[0015] S404. According to the above scores and construct the loss function: For the CNN discriminator, the loss function is ; for the UNet generator, the loss function is 。

[0016] S405. Repeat steps S401 - S404. For all samples, accumulate the loss function: , S406. According to the above loss function, use the gradient descent method to train the two networks: , 。

[0017] S407. Repeat steps S401 - S406 to train the two networks repeatedly until convergence. The dataset for each process is the same. The dataset is loaded using mini - batch, batchsize = 20, and the current training stop strategy is to train for 50 epochs; at this time, the UNet generator can accurately estimate the radio map based on the user's received signal strength, location, and geographical environment map information; and when there are certain errors in the geographical environment map information, it can automatically correct some errors and still ensure a high estimation accuracy. The specific effects are as Figures 4 - 5 shown. Figure 4 The estimation accuracy of the proposed scheme when the physical environment map has no error: Among them, (left) label, (middle) estimation effect based on 300 RSS sampling points, (right) estimation effect based on 1000 RSS sampling points; Figure 5 The estimation accuracy of the proposed scheme when the physical environment map has errors (some buildings are missing, the blackened part): (left) label, (middle) estimation effect based on 300 RSS sampling points, (right) estimation effect based on 1000 RSS sampling points S5. In the inference stage, use the generator to estimate the radio map: that is, obtain the received signal strength information, RSS information, and geographical environment information of multiple users as data features, input them into the generator, and the generator outputs the radio map estimation result.

[0018] In summary, this solution does not require transmitter information. Instead, it uses the received signal strength and location information of the user terminal, combined with environmental information, and inputs them into a deep neural network to obtain a radio map estimate. Considering that in a dynamic wireless network, the transmitter may be any base station, drone, or mobile user, and it is often difficult to obtain transmitter information in real time. On the other hand, the received signal strength and location information of the user terminal can be obtained in real time through the user's uplink, which has strong practicality. At the same time, the traditional method of generating a radio map using a transmitter relies on the physical model of signal propagation. When there is an error in the transmitter information, it will inevitably lead to an error in the radio map estimate, and according to the signal propagation characteristics, there is a certain amplification effect. The present invention is based on the received signal strength information of the user terminal, which is an accurate measured result, so this problem does not exist. At the same time, in this solution, the distributed received signal strength information of the user terminal can be regarded as sparse sampling on the real radio map. As the number of users increases, the available data information also increases, and the accuracy of the radio map estimate continues to improve. Currently, there are often certain errors in maps, including missing buildings, temporary obstacles such as pedestrians or vehicles, and missing environmental details. In this solution, the received signal strength of the user terminal is itself the result of the interaction between the transmitter signal and the surrounding environment. Therefore, the received signal strength information of the user terminal contains the deep features of the surrounding environment. When the number of users is sufficient (for example, there are a certain number of users around a missing building), the deep neural network can extract the deep features in the received signal strength information of the users, correct the errors in the environmental information, obtain a relatively accurate radio map, and at the same time give the corrected environmental information.

[0019] The above is the preferred implementation manner of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, should not be regarded as excluding other embodiments, and can be used in other combinations, modifications, and environments, and can be changed within the scope of the concept described herein through the above teachings or the technology or knowledge in related fields. And the changes and modifications made by those skilled in the art without departing from the spirit and scope of the present invention should all be within the protection scope of the appended claims of the present invention.

Claims

1. A radio map collaborative estimation method based on a generative adversarial network, characterized by: The following steps are involved: S1. Use the received signal strength information, RSS information and geographic environment information of multiple users as data features and the radio map as a label to construct a training data set, where the RSS information is the location information; S2. The UNet model is used as the generator, with the input being the graphical geographical environment map and RSS information, and the output being the graphical radio map estimation result; S3. A convolutional neural network model, namely a CNN model, is used as a discriminator to evaluate the authenticity of the radio map output by the generator; S4. During the model training phase, the generator and discriminator are jointly optimized; S5. In the inference phase, the generator is used for radio map estimation.

2. The method for collaboratively estimating a radio map based on a generative adversarial network according to claim 1, characterized in that: In step S1, it is assumed that the training data set contains L data samples. Data samples: The data features are ,in Represents the matrix composed of the received signal strength information and RSS information of all users, is a matrix representing geographic environment information. express The corresponding radio map, i.e. The labels of the data samples, where .

3. The method for collaboratively estimating a radio map based on a generative adversarial network according to claim 1, characterized in that: In step S2, the UNet network is used as the generator, which is expressed as , whose parameter matrix is .

4. The method for collaboratively estimating a radio map based on a generative adversarial network according to claim 1, characterized in that: In step S3, a CNN network is selected as the discriminator, which is expressed as , whose parameter matrix is .

5. The method for collaboratively estimating a radio map based on a generative adversarial network according to claim 3, characterized in that: The step S4 comprises: S401. For any data sample , the data features Input to the UNet generator to calculate the "false label" ; S402. Pair data features with true labels And input to the CNN discriminator to calculate the score ; S403. Pairing data features with "false labels" And input to the CNN discriminator to calculate the score ; S404. Based on the above scores and Construct loss function: For CNN discriminator, the loss function is ; For the UNet generator, the loss function is ; represents the 2-norm; S405. Repeat steps S401 to S404 for all samples and accumulate the loss function: , ; S406. Use gradient descent to train the UNet generator and CNN discriminator: , ; S407. Carry out the next round of training. The training process repeats S401 to S406 until the set training round is reached.

Images