Radio map reconstruction method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of wireless communication, and provides a radio map reconstruction method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining the environment information and sparse measurement of a target region; inputting the environment information into a trained prediction network to obtain a basic radio map output by the prediction network; inputting the basic radio map and the sparse measurement into a trained correction network to obtain a precise radio map output by the correction network; wherein the prediction network and the correction network are both obtained based on adversarial training between the generator and the discriminator. According to the radio map reconstruction method and device, the electronic equipment and the storage medium provided by the invention, the optimal result can be obtained by utilizing the existing available information, so that the balance among efficiency, cost and precision is realized.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to a radio map reconstruction method, device, electronic equipment and storage medium. Background Art

[0002] With the development of next-generation wireless communication systems, reliable and efficient channel models are increasingly needed to reflect the status of wireless signals under different conditions. One of the effective technologies is to construct a radio map, that is, to use images to represent communication quality indicators such as average received signal power under the superposition of concurrent wireless transmissions. Traditional radio maps can be constructed through empirical models, such as obtaining the received signal power of each receiving point through path loss models such as COST231-Hata. This method is simple to calculate and fast, but the result is uniform and isotropic, and the specific propagation environment is not considered, and it cannot be accurately predicted in complex environments. In order to improve the accuracy, deterministic models represented by ray tracing can use the law of signal propagation to simulate and predict based on a given three-dimensional environment model, but the disadvantage of this method is that the calculation cost is high and it takes a long time.

[0003] In order to meet the requirements of both accuracy and speed, the latest research focuses on applying machine learning (ML) and deep learning (DL) techniques to the construction of radio maps. Existing methods can be classified according to the required input information. The first type of method is to interpolate based on discrete measurements to restore the complete radio map. In order to ensure sufficient accuracy, a large amount of measurement data that can cover most areas must be collected, which requires a lot of manpower and time costs. The second type is a method based on prediction based on prior environmental information, which can be obtained in advance through other means without real-time measurement. Commonly used environmental information includes transmitter configuration and obstacle information in the area, such as obstacle location, obstacle height, top view and side view of the area. Due to the limited environmental information obtained, it is impossible to accurately represent all factors related to signal propagation, so the prediction accuracy is usually low. The third type of method uses both measurement and environmental information and takes them as input. Although the advantages of both types of information can be used to increase accuracy, the time consistency of the input information is not considered, which greatly reduces the efficiency. Summary of the invention

[0004] The present invention provides a radio map reconstruction method, device, electronic device and storage medium, which are used to solve the defect that radio map construction in the prior art is difficult to achieve a balance between accuracy, cost and efficiency.

[0005] The present invention provides a radio map reconstruction method, comprising: Obtain environmental information and sparse measurements of the target area; Inputting the environmental information into a trained prediction network to obtain a basic radio map output by the prediction network; Inputting the basic radio map and the sparse measurement into a trained correction network to obtain an accurate radio map output by the correction network; Wherein, the prediction network and the correction network are both obtained based on adversarial training between the generator and the discriminator.

[0006] According to the radio map reconstruction method provided by the present invention, the step of inputting the environmental information into a trained prediction network to obtain a basic radio map output by the prediction network comprises: Based on the transmitter information and obstacle information in the environmental information, an empirical channel map and a line-of-sight or non-line-of-sight indication map are generated through an empirical path loss model; The empirical channel map, the line-of-sight or non-line-of-sight indication map, the transmitter position in the transmitter information, and the obstacle overhead view in the obstacle information are input into a trained prediction network to obtain a basic radio map output by the prediction network.

[0007] According to the radio map reconstruction method provided by the present invention, the empirical channel map, the line-of-sight or non-line-of-sight indication map, the transmitter position in the transmitter information, and the obstacle top view in the obstacle information are input into a trained prediction network to obtain a basic radio map output by the prediction network, including: Inputting the empirical channel map, the line-of-sight or non-line-of-sight indication map, the transmitter position and the obstacle top view into the feature extraction layer of the prediction network to obtain a first input feature; Inputting the first input feature into the self-attention layer of the prediction network to obtain an attention feature; The attention features are input into a prediction layer of the prediction network to obtain the basic radio map.

[0008] According to the radio map reconstruction method provided by the present invention, the line-of-sight or non-line-of-sight indication map is generated based on the relative positions among the transmitter, the receiver and the obstacles in the target area.

[0009] According to the radio map reconstruction method provided by the present invention, the basic radio map and the sparse measurement are input into a trained correction network to obtain an accurate radio map output by the correction network, comprising: Based on the radial basis function interpolation method, the sparse measurement is interpolated to obtain an interpolation result; The basic radio map, the sparse measurement and the interpolation result are input into a trained correction network to obtain an accurate radio map output by the correction network.

[0010] According to the radio map reconstruction method provided by the present invention, the basic radio map, the sparse measurement and the interpolation result are input into a trained correction network to obtain an accurate radio map output by the correction network, comprising: Inputting the base radio map, the sparse measurement and the interpolation result into a feature extraction layer of the correction network to obtain a second input feature, and extracting an intermediate feature of the second input feature; Inputting the second input feature and the intermediate feature into the residual connection layer of the correction network to obtain residual enhanced features; The residual enhancement feature is input into a correction layer of the correction network to obtain the precise radio map output by the correction layer.

[0011] According to the radio map reconstruction method provided by the present invention, the loss functions of the prediction network and the correction network both include adversarial loss and mean square error loss.

[0012] The present invention also provides a radio map reconstruction device, comprising: An acquisition unit, used to acquire environmental information and sparse measurements of a target area; A prediction unit, configured to input the environmental information into a trained prediction network to obtain a basic radio map output by the prediction network; A correction unit, configured to input the basic radio map and the sparse measurement into a trained correction network to obtain a precise radio map output by the correction network; Wherein, the prediction network and the correction network are both obtained based on adversarial training between the generator and the discriminator.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any one of the radio map reconstruction methods described above is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the radio map reconstruction method as described above is implemented.

[0015] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned radio map reconstruction methods.

[0016] The radio map reconstruction method, device, electronic device and storage medium provided by the present invention obtain the environmental information and sparse measurement of the target area; input the environmental information into the trained prediction network to obtain the basic radio map output by the prediction network; then input the basic radio map and sparse measurement into the trained correction network to obtain the accurate radio map output by the correction network. The best results can be obtained by using the existing available information, thereby achieving a balance between efficiency, cost and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 It is a flow chart of the radio map reconstruction method provided by the present invention.

[0019] Figure 2 It is a schematic diagram of the structure of the generator and the discriminator provided by the present invention.

[0020] Figure 3 It is a structural schematic diagram of the self-attention layer provided by the present invention.

[0021] Figure 4 It is a structural schematic diagram of the residual connection layer provided by the present invention.

[0022] Figure 5 It is a comparison chart of the generated results provided by the present invention.

[0023] Figure 6 These are visualized radio maps in three typical scenarios provided by the present invention.

[0024] Figure 7 It is the average performance of RMC-GAN under different proportions of measurement values ​​provided by the present invention.

[0025] Figure 8 It is a structural schematic diagram of the radio map reconstruction device provided by the present invention.

[0026] Fig. 9 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] In view of the problem that it is difficult to achieve a balance between accuracy, cost and efficiency in radio map construction in the prior art, an embodiment of the present invention provides a radio map construction architecture of prediction first and correction later (FPTC). In this method, environmental information and sparse measurements of the target area are first obtained; the environmental information is input into a trained prediction network to obtain a basic radio map output by the prediction network; the basic radio map and sparse measurements are then input into a trained correction network to obtain an accurate radio map output by the correction network. The best results can be obtained by using the existing available information, thereby achieving a balance between efficiency, cost and accuracy.

[0029] The embodiments of the present invention can be applied to scenarios where radio map reconstruction is required. The execution subject of the method can be an electronic device such as a terminal device, a computer, a server, a server cluster, or a specially designed radio map reconstruction device, or a radio map reconstruction device set in the electronic device, which can be implemented by software, hardware, or a combination of the two.

[0030] In the description of the embodiments of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0031] Figure 1 is a flow chart of the radio map reconstruction method provided by the present invention, such as Figure 1 As shown, the method comprises the following steps: Step 110, obtaining environmental information and sparse measurements of the target area; Step 120, inputting the environmental information into the trained prediction network to obtain a basic radio map output by the prediction network; Step 130, inputting the basic radio map and the sparse measurement into the trained correction network to obtain the accurate radio map output by the correction network; Among them, the prediction network and the correction network are trained based on the sample environment information and its corresponding labeled radio map.

[0032] Specifically, the target area is the area where the radio map needs to be constructed. The environmental information of the target area may include transmitter information and obstacle information. There are usually multiple transmitters and obstacles of different densities The target area is usually a small part of it, defined as , contains only a limited number of emitters and obstacles .

[0033] However, due to the characteristics of signal propagation, the received signal in the target area will be affected by the external unknown environment. The actual radio map P is thus expressed as: (1)

[0034] in, Indicates based on The channel modeling formula is: The impact caused by the unknown external environment.

[0035] The complete environmental information in the ,contains multiple aspects, such as crowds and weather.,In order to reduce costs, only the relatively easy to obtain in advance and the small changes in a short time, and As the environmental information is available, in order to better model , using a deep learning network To predict a rough radio map , expressed as: (2)

[0036] In existing work, It is often ignored as an ideal, assuming that the target area is completely independent, which is inconsistent with reality. Unpredictable, can use sparse measurements As a correction network Input to correct Make it closer to the real radio map P. In addition, due to other factors (such as The prediction error caused by environmental factors not considered in the calculation can also be corrected by this step. It is expressed as:

[0037] Here, the environmental information of the target area can be acquired in advance, and the sparse measurements must be collected through field measurement activities. To ensure the temporal consistency of the input, the present invention proposes a prediction-first-correction-later (FPTC) architecture, which is implemented through two sub-networks with different structures, namely a prediction network RMP-GAN and a correction network RMC-GAN, which are used to process environmental information and sparse measurements respectively. This method is called FPTC-GANs. The input of the prediction network is environmental information, and the output is a rough radio map. , which is the basic radio map. The input of the correction network is the basic radio map and sparse measurements, and the correction network outputs a more accurate radio map , i.e., precise radio maps.

[0038] It should be noted that if there is no measurement data, the predicted basic radio map can be directly used for basic applications, such as drone path planning, base station deployment, resource allocation, security monitoring, etc. The advantage of this method is that the best results can be obtained using the existing available information, thus achieving a balance between efficiency, cost and accuracy.

[0039] In some embodiments, the environment information is input into a trained prediction network to obtain a basic radio map output by the prediction network, that is, step 120 specifically includes: Step 121, based on the transmitter information and obstacle information in the environmental information, generate an empirical channel map and a line-of-sight or non-line-of-sight indication map through an empirical path loss model; Step 122, input the empirical channel map, the line-of-sight or non-line-of-sight indication map, the transmitter position in the transmitter information, and the obstacle top view in the obstacle information into the trained prediction network to obtain a basic radio map output by the prediction network.

[0040] Specifically, in order to facilitate the prediction of the network Learn the mapping relationship between input and output, and transform the target area Rasterizes to a 2D image format to represent input and output. Transmitter Information Including transmitter location , height, transmission power and frequency; obstacle information Includes top view of obstacles Combining the transmitter information with the obstacle information yields a radio map based on the empirical path loss model COST231-Hata. , that is, empirical channel maps, and line-of-sight or non-line-of-sight indication maps .

[0041] Then, the experience channel map , line-of-sight or non-line-of-sight indication diagram , transmitter location in transmitter information , obstacle top view in obstacle information Input to the trained prediction network to obtain the basic radio map output by the prediction network .

[0042] Generating an Experience Channel Map Considering that a large amount of training data is usually used in the related art to calibrate the empirical path loss model, although more accurate parameter values ​​can be obtained, the generalization is greatly reduced. In order to enhance the generalization, the embodiment of the present invention does not further calibrate the empirical path loss model COST231-Hata, but directly uses the original parameters. The path loss PL is expressed as: (3)

[0043] in is the frequency in MHz, is the height of the transmitter above the ground in metres, is the distance between the transmitter and the receiver in kilometers. In an urban scenario, is 3dB, ,in Height of the receiver above the ground in meters.

[0044] Here, the line-of-sight or non-line-of-sight indicator It is generated based on the relative positions of the transmitter, receiver, and obstacles. In, if Then the point is line of sight, otherwise it is non-line of sight, where Indicates the connection between the transmitter and the receiver Previous The height of the point, Represents the boundary value of the line-of-sight and non-line-of-sight paths calculated using the geometric method. The calculation formula is: (4)

[0045] in and Represents the height of the transmitter and receiver, the transmitter, the receiver and the line between them Previous The locations of the points are , and .

[0046] Based on any of the above embodiments, the empirical channel map, the line-of-sight or non-line-of-sight indication map, the transmitter position in the transmitter information, and the obstacle top view in the obstacle information are input into the trained prediction network to obtain a basic radio map output by the prediction network, specifically including: Step 122-1, inputting the empirical channel map, the line-of-sight or non-line-of-sight indication map, the transmitter location and the obstacle top view into the feature extraction layer of the prediction network to obtain a first input feature; Step 122-2, inputting the first input feature into the self-attention layer of the prediction network to obtain an attention feature; Step 122-3, input the attention features into the prediction layer of the prediction network to obtain the basic radio map.

[0047] Specifically, the prediction network and the correction network have in common that both are based on adversarial training between the generator and the discriminator. The generator generates a radio map based on the input, and the discriminator distinguishes between true and false. The difference between the two networks is that, based on the different prediction / correction tasks, corresponding functional modules are additionally designed on the basis of the original CGAN, namely, a self-attention (SA) module is designed for the prediction network, and a residual connection (RC) module is designed for the correction network.

[0048] Preferably, the structure of the generator is the same as that of UNet, with downsampling and upsampling performed successively. The downsampling layer consists of a convolutional layer (Conv), batch normalization (BN), and an activation layer (Leaky ReLu). The upsampling layer consists of a deconvolutional layer (ConvT), a Dropout layer (Drop), and an activation layer. The discriminator uses the downsampling structure to extract features to distinguish authenticity. Figure 2 It is a schematic diagram of the structure of the generator and the discriminator provided by the present invention, such as Figure 2 As shown, (SA) p Indicates that the corresponding self-attention module SA is only used for prediction network, (RC) c It means that the corresponding residual module RC is only used to correct the network.

[0049] In this embodiment, a self-attention layer (SA) is added to the prediction network. Figure 3 Schematic diagram of the structure of the self-attention layer provided by the present invention. Radio map prediction involves understanding the spatial dependencies between distant locations, and the self-attention mechanism helps capture global relationships and focus on important features. Given the first input feature , the self-attention layer can enhance feature capture, expressed as: (5)

[0050] in , and For passing The convolved query, key, and value, and Used to calculate the attention score through softmax, and the calculation result is the same as Multiply to get the weighted result , usage coefficient After scaling Add and get the output , that is, the attention feature.

[0051] Then, the attention features are input into the prediction layer of the prediction network to obtain the basic radio map.

[0052] Based on any of the above embodiments, the basic radio map and the sparse measurement are input into the trained correction network to obtain the accurate radio map output by the correction network, that is, step 130 specifically includes: Step 131, based on the radial basis function interpolation method, the sparse measurement is interpolated to obtain an interpolation result; Step 132, input the basic radio map, sparse measurement and interpolation results into the trained correction network to obtain the accurate radio map output by the correction network.

[0053] Specifically, sparse measurement Integrated into , generated using radial basis function (RBF) interpolation Sparse Measurement , interpolation results and basic radio maps As input to the correction network, the correction network outputs a refined radio map.

[0054] Interpolation results Using RBF interpolation method, each location without data Value It is expressed as: (6)

[0055] in is the number of known points, and Respectively for The positions and weights of the known points, is the radial basis function, which depends on the distance between two points .

[0056] The base radio map, sparse measurements, and interpolation results are fed into the trained correction network to obtain an accurate radio map output by the correction network, including: Step 132-1, inputting the basic radio map, the sparse measurement and the interpolation result into the feature extraction layer of the correction network to obtain the second input feature, and extracting the intermediate feature of the second input feature; Step 132-2, inputting the second input feature and the intermediate feature into the residual connection layer of the correction network to obtain the residual enhancement feature; Step 132 - 3 , inputting the residual enhancement features into the correction layer of the correction network to obtain the accurate radio map output by the correction layer.

[0057] Specifically, the revised network also includes a generator and a discriminator, and a residual connection layer is added in this embodiment. Figure 4 is a schematic diagram of the structure of the residual connection layer provided by the present invention. In the residual connection layer, the second input feature After Conv, BN and ReLu processing, the intermediate features are obtained. The intermediate features are consistent with the original second input features. The residual connection layer can retain more valuable input information, allowing the model to focus on completing the correction task instead of rebuilding the entire radio map from scratch. The enhanced features are is the intermediate feature after Conv, BN and activation (Relu), and After adding, the activation is output .

[0058] In some embodiments, the loss functions of the prediction network and the correction network both include adversarial loss and mean square error loss.

[0059] Specifically, the prediction network and the correction network provided by the present invention use the same target true value to calculate the target true value and or The loss function includes adversarial loss and mean square error loss. The adversarial loss is: (7)

[0060] in and Input And the distribution of the target true value y. The mean square error loss is: (8)

[0061] The method provided in the embodiment of the present invention proposes an FPTC architecture, which includes a prediction network and a correction network, and can use the existing available information to obtain the best results, thereby achieving a balance between efficiency, cost and accuracy. In view of the task characteristics of the prediction network, a network structure including a self-attention layer is designed to improve the prediction accuracy. At the same time, a new environmental representation method of the line-of-sight and non-line-of-sight indicator map is used to facilitate network understanding, and the uncorrected original COST231-Hata model is used as a guide to enhance generalization. In view of the characteristics of the correction network, a network structure including a residual connection layer is designed to improve the prediction accuracy, and the RBF interpolation method is used to assist the network in correcting local trends.

[0062] In order to further verify the effect of the radio map reconstruction method provided by the present invention, the experimental results are as follows: (1) Dataset

[0063] According to the required input and expected results, the BART-Lab radio map dataset can be used. In this dataset, each radio map contains three transmitters at the same time. The dataset contains a total of 2000 radio maps, each of which has a corresponding top view of obstacles and transmitter location information, and the image size ranges from 140×268 pixels to 561×1313 pixels. In order to simulate the required target area, the image is cut into a fixed size of 128×128 pixels, and one transmitter and some obstacles are retained on the same radio map. For sparse sampling, 120 points (about 0.73%) are randomly selected from the non-building part of each radio map. After preprocessing, a total of 3518 sets of valid data are obtained, and the obstacle density, distribution and transmitter location in each set of data are different, which greatly improves the generalization ability of the model. These data are divided into two training sets, two validation sets and one test set to train the two networks and compare their performance. The data volume ratio of the five sub-datasets is 4:3:1:1:1.

[0064] (2) Network performance

[0065] The experimental results are shown in Table 1. First, the effectiveness of the SA module can be verified by comparing RMP-GAN and CGAN in the prediction task, while the comparison results of RMC-GAN and CGAN in the correction task can prove the effectiveness of the RC module (for fair comparison, the basic radio map inputs used by the two networks in the correction task are all generated by RMP-GAN). RMP-GAN and RMC-GAN both show better performance than CGAN in the corresponding tasks, while the computational complexity only increases slightly to an acceptable degree.

[0066] Secondly, following the proposed FPTC framework, the correction step is performed after the prediction step. The effectiveness of this framework can be verified by comparing the results of RMP-GAN and RMC-GAN. The results show that the accuracy after correction is significantly improved in four dimensions (RMSE, MAE, SSIM and PSNR).

[0067] At the same time, the method proposed in the present invention is compared with other existing methods. Among them, the network model of RadioUNet adopts WNet; GAN-CRME uses the original GAN ​​through the cooperative radio map estimation (CRME) method; Pix2Pix+ trains the model to learn the difference between the real radio map and the empirical model based on the Pix2Pix network. However, in order to ensure the fairness of the comparison, the uncorrected empirical model is used in the experiment, so the performance of Pix2Pix+ may be slightly impaired; the first three methods use environmental information and sparse measurements at the same time, but none of them follow the FPTC framework and do not consider the temporal consistency of the input; DeepREM interpolates only based on the measured values. In the experiment, the number of measured values ​​is consistent with the original text (5%), which is significantly higher than the number used by the method proposed in the present invention (0.73%). The results show that the FPTC-GANs method proposed in the present invention is superior to other methods in terms of accuracy. Although the computational complexity is slightly higher due to the existence of the second network and additional functional modules, the increment is acceptable compared with the improvement in accuracy, which proves the effectiveness of the method.

[0068] Table 1 Average performance of different models in the test set

[0069] In addition, RMP-GAN performs significantly better than CGAN in some scenarios. Figure 5 is a comparison chart of the generated results provided by the present invention, such as Figure 5 As shown in the figure, the RMSE comparison of the generated results of CGAN and RMP-GAN in some specific scenarios shows that the RMSE of each scenario is reduced by more than 2.3 dBm, which further verifies the effectiveness of RMP-GAN.

[0070] Figure 6It is a visualized radio map in three typical scenarios provided by the present invention, wherein scenario 1 contains obstacles with low density, scenario 2 has densely distributed obstacles, and scenario 3 is significantly affected by the external environment. The following uses these scenarios as examples to illustrate the observation results. For the prediction network, the area with higher obstacle density usually has lower prediction accuracy, for example, the accuracy of scenario 1 is higher than that of scenario 2. This is because the signal propagation law in the area with sparse obstacles is relatively simple, which is basically consistent with the empirical model in free space and is easier for the model to learn; as the density of obstacles increases, the propagation rules involved are more complex, resulting in reduced accuracy. In addition, the higher the degree of influence of the external environment, the greater the prediction error. In scenario 3, the upper left corner and the upper right corner are greatly affected by the nearby transmitters, resulting in a large error. This is because only the environmental information in the target area is available, and it is difficult for the model to learn relevant information about the unknown environment. For the correction network, the larger the prediction error, the greater the degree of improvement in accuracy after correction. For example, the error reduction of scenario 3 with a large error after correction is significantly higher than that of the first two scenarios. Therefore, the correction step is particularly important for scenarios with low prediction accuracy.

[0071] (3) Ablation experiment

[0072] Since each network contains multiple inputs, it is necessary to verify the role of each input. According to the idea of ​​ablation study, the experimental results are shown in Table 2. For RMP-GAN, we first verify whether the empirical model and line-of-sight and non-line-of-sight information can provide useful guidance for the training process. Therefore, we verify the removal of and The impact on the results. At the same time, due to The approximate location of the transmitter in the target area is also implied, so it is necessary to verify It is not redundant information. For RMC-GAN, it is necessary to verify the RBF interpolation results. The results show that the above ablation experiments have a decrease in accuracy and image quality compared to the results of the original input. Therefore, it can be proved that all the above inputs are helpful in building accurate radio maps.

[0073] Table 2 Average performance of models with different inputs in the test set

[0074] also, Figure 7 is the average performance of RMC-GAN under different ratios of measurements provided by the present invention. From the comparison results, it can be seen that the more the number of measurements, the higher the accuracy and image quality of the correction result. However, collecting more measurements requires more cost. In practical applications, it is necessary to balance accuracy and cost to determine the appropriate number of measurements.

[0075] The radio map reconstruction device provided by the present invention is described below. The radio map reconstruction device described below and the radio map reconstruction method described above can be referred to each other.

[0076] Based on any of the above embodiments, Figure 8 is a schematic diagram of the structure of the radio map reconstruction device provided by the present invention, such as Figure 8 As shown, the device comprises: An acquisition unit 810, configured to acquire environmental information and sparse measurements of a target area; A prediction unit 820, configured to input the environmental information into a trained prediction network to obtain a basic radio map output by the prediction network; A correction unit 830, configured to input the basic radio map and the sparse measurement into a trained correction network to obtain a precise radio map output by the correction network; Wherein, the prediction network and the correction network are both obtained based on adversarial training between the generator and the discriminator.

[0077] The device provided by the embodiment of the present invention can obtain the best result by using the existing available information, thereby achieving a balance between efficiency, cost and accuracy.

[0078] Based on any of the above embodiments, the prediction unit is specifically used for: Based on the transmitter information and obstacle information in the environmental information, an empirical channel map and a line-of-sight or non-line-of-sight indication map are generated through an empirical path loss model; The empirical channel map, the line-of-sight or non-line-of-sight indication map, the transmitter position in the transmitter information, and the obstacle overhead view in the obstacle information are input into a trained prediction network to obtain a basic radio map output by the prediction network.

[0079] Based on any of the above embodiments, the prediction unit is specifically used for: Inputting the empirical channel map, the line-of-sight or non-line-of-sight indication map, the transmitter position and the obstacle top view into the feature extraction layer of the prediction network to obtain a first input feature; Inputting the first input feature into the self-attention layer of the prediction network to obtain an attention feature; The attention features are input into a prediction layer of the prediction network to obtain the basic radio map.

[0080] Based on any of the above embodiments, the line-of-sight or non-line-of-sight indication map is generated based on the relative positions of the transmitter, the receiver and the obstacles in the target area.

[0081] Based on any of the above embodiments, the correction unit is specifically used for: Based on the radial basis function interpolation method, the sparse measurement is interpolated to obtain an interpolation result; The basic radio map, the sparse measurement and the interpolation result are input into a trained correction network to obtain an accurate radio map output by the correction network.

[0082] Based on any of the above embodiments, the correction unit is specifically used for: Inputting the base radio map, the sparse measurement and the interpolation result into a feature extraction layer of the correction network to obtain a second input feature, and extracting an intermediate feature of the second input feature; Inputting the second input feature and the intermediate feature into the residual connection layer of the correction network to obtain residual enhanced features; The residual enhancement feature is input into a correction layer of the correction network to obtain the precise radio map output by the correction layer.

[0083] Based on any of the above embodiments, the loss functions of the prediction network and the correction network both include adversarial loss and mean square error loss.

[0084] Fig. 9 An example of a physical structure diagram of an electronic device is shown in FIG. Fig. 9 As shown, the electronic device may include: a processor 910, a communication interface 920, a memory 930 and a communication bus 940, wherein the processor 910, the communication interface 920 and the memory 930 communicate with each other through the communication bus 940. The processor 910 may call the logic instructions in the memory 930 to execute the radio map reconstruction method, which includes: obtaining environmental information and sparse measurements of the target area; inputting the environmental information into a trained prediction network to obtain a basic radio map output by the prediction network; inputting the basic radio map and the sparse measurements into a trained correction network to obtain an accurate radio map output by the correction network; wherein the prediction network and the correction network are both obtained based on adversarial training between a generator and a discriminator.

[0085] In addition, the logic instructions in the above-mentioned memory 930 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0086] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the radio map reconstruction method provided by the above methods, which method includes: obtaining environmental information and sparse measurements of the target area; inputting the environmental information into a trained prediction network to obtain a basic radio map output by the prediction network; inputting the basic radio map and the sparse measurements into a trained correction network to obtain an accurate radio map output by the correction network; wherein the prediction network and the correction network are both obtained based on adversarial training between a generator and a discriminator.

[0087] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the radio map reconstruction method provided by the above-mentioned methods, the method comprising: obtaining environmental information and sparse measurements of the target area; inputting the environmental information into a trained prediction network to obtain a basic radio map output by the prediction network; inputting the basic radio map and the sparse measurements into a trained correction network to obtain an accurate radio map output by the correction network; wherein the prediction network and the correction network are both obtained based on adversarial training between a generator and a discriminator.

[0088] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0089] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A radio map reconstruction method, characterized in that: include: Obtain environmental information and sparse measurements of the target area; Inputting the environmental information into a trained prediction network to obtain a basic radio map output by the prediction network; Inputting the basic radio map and the sparse measurement into a trained correction network to obtain an accurate radio map output by the correction network; Wherein, the prediction network and the correction network are both obtained based on adversarial training between the generator and the discriminator.

2. The radio map reconstruction method according to claim 1, characterized in that: The step of inputting the environmental information into a trained prediction network to obtain a basic radio map output by the prediction network comprises: Based on the transmitter information and obstacle information in the environmental information, an empirical channel map and a line-of-sight or non-line-of-sight indication map are generated through an empirical path loss model; The empirical channel map, the line-of-sight or non-line-of-sight indication map, the transmitter position in the transmitter information, and the obstacle overhead view in the obstacle information are input into a trained prediction network to obtain a basic radio map output by the prediction network.

3. The radio map reconstruction method according to claim 2, characterized in that: The step of inputting the empirical channel map, the line-of-sight or non-line-of-sight indication map, the transmitter position in the transmitter information, and the obstacle overhead view in the obstacle information into a trained prediction network to obtain a basic radio map output by the prediction network includes: Inputting the empirical channel map, the line-of-sight or non-line-of-sight indication map, the transmitter position and the obstacle top view into the feature extraction layer of the prediction network to obtain a first input feature; Inputting the first input feature into the self-attention layer of the prediction network to obtain an attention feature; The attention features are input into a prediction layer of the prediction network to obtain the basic radio map.

4. The radio map reconstruction method according to claim 2, characterized in that: The line-of-sight or non-line-of-sight indication map is generated based on the relative positions of the transmitter, the receiver and obstacles in the target area.

5. The radio map reconstruction method according to claim 1, characterized in that: The step of inputting the basic radio map and the sparse measurement into a trained correction network to obtain an accurate radio map output by the correction network comprises: Based on the radial basis function interpolation method, the sparse measurement is interpolated to obtain an interpolation result; The basic radio map, the sparse measurement and the interpolation result are input into a trained correction network to obtain an accurate radio map output by the correction network.

6. The radio map reconstruction method according to claim 5, characterized in that: The step of inputting the basic radio map, the sparse measurement and the interpolation result into a trained correction network to obtain an accurate radio map output by the correction network comprises: Inputting the base radio map, the sparse measurement and the interpolation result into a feature extraction layer of the correction network to obtain a second input feature, and extracting an intermediate feature of the second input feature; Inputting the second input feature and the intermediate feature into the residual connection layer of the correction network to obtain residual enhanced features; The residual enhancement feature is input into a correction layer of the correction network to obtain the precise radio map output by the correction layer.

7. The radio map reconstruction method according to any one of claims 1 to 6, characterized in that: The loss functions of the prediction network and the correction network both include adversarial loss and mean square error loss.

8. A radio map reconstruction device, characterized in that: include: An acquisition unit, used to acquire environmental information and sparse measurements of a target area; A prediction unit, configured to input the environmental information into a trained prediction network to obtain a basic radio map output by the prediction network; A correction unit, configured to input the basic radio map and the sparse measurement into a trained correction network to obtain a precise radio map output by the correction network; Wherein, the prediction network and the correction network are both obtained based on adversarial training between the generator and the discriminator.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the radio map reconstruction method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the radio map reconstruction method according to any one of claims 1 to 7 is implemented.

Images