Environment feature assisted channel multipath intelligent prediction method and device

Abstract

The application provides an environment feature assisted channel multipath intelligent prediction method and device, which comprises the following steps: inputting channel multipath component geometric information and channel multipath component field strength information of a plurality of historical time moments of a terminal to be predicted, coordinate information corresponding to the terminal to be predicted at the plurality of historical time moments, respectively, and scene geometric features of a wireless propagation environment in which the terminal to be predicted is located, and inputting coordinate information and material electromagnetic parameter features of the wireless propagation environment into a feature extraction subnetwork of a multipath long-period prediction network, respectively, to obtain three feature maps; inputting the three feature maps into a sequence prediction subnetwork of the multipath long-period prediction network to obtain channel multipath component prediction information of a plurality of future time moments of the terminal to be predicted. In this embodiment, a network for a time sequence prediction task of channel multipaths is built, environment features are introduced, and the influence of the environment on electromagnetic wave propagation is mined, so that the prediction accuracy is further improved, and the long-period prediction capability is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication, in particular to an environment feature assisted channel multipath intelligent prediction method and device. BACKGROUND

[0002] The sixth-generation (6G) mobile communication system is expected to provide intelligent, ultra-reliable and ubiquitous connectivity, supporting the Internet of Everything (IoE) trend of the future digital society. However, the global coverage and diversified scenarios brought by 6G bring extreme complexity and new challenges to network design and technology optimization. Digital twin technology is considered an innovative tool to overcome these challenges. The mapping of the propagation behavior of electromagnetic waves and its physical layer operation in the digital world is defined as a digital twin channel (DTC). The DTC is expected to have the ability to predict future channel changes, and can promote the optimization of various tasks in the network, early warning and decision-making, such as network planning, resource allocation and link operation, and realize active adaptation to the complex wireless environment of 6G.

[0003] With the development of artificial intelligence (AI), AI has been widely explored for intelligent prediction of wireless channels. However, in AI-based channel prediction, current research still focuses on predicting large-scale parameters of the channel, such as path loss, delay spread, etc., and lacks methods for predicting the multipath components (MPCs) of the channel at future time points, including the geometric information and field strength information of the MPCs. SUMMARY

[0004] The purpose of the present application is to provide an environment feature assisted channel multipath intelligent prediction method and device, which solves the problem that in the prior art, in AI-based channel prediction, still focuses on predicting large-scale parameters of the channel, such as path loss, delay spread, etc., and lacks methods for predicting the MPCs at future time points, including the geometric information and field strength information of the MPCs.

[0005] To achieve the above purpose, the embodiments of the present application provide an environment feature assisted channel multipath intelligent prediction method, which comprises:

[0006] The channel multipath component geometric information and channel multipath component field strength information of a plurality of historical time points of a terminal to be predicted are input into a first feature extraction subnetwork of a multipath long-period prediction network to obtain a first feature map; wherein the first feature map includes the relationship between the geometry and the field strength of the channel multipath component of the terminal to be predicted;

[0007] input the coordinate information corresponding to the plurality of historical time instants of the terminal to be predicted and the scene geometric features of the wireless propagation environment in which the terminal to be predicted is located into a second feature extraction subnetwork of the multipath long-period prediction network to obtain a second feature map; wherein the second feature map comprises the relationship between the geometric features and the scene geometric features of the channel multipath components;

[0008] input the coordinate information and the material electromagnetic parameter features of the wireless propagation environment into a third feature extraction subnetwork of the multipath long-period prediction network to obtain a third feature map; wherein the third feature map comprises the relationship between the field intensity of the channel multipath components and the scene material electromagnetic parameter features;

[0009] input the first feature map, the second feature map and the third feature map into a sequence prediction subnetwork of the multipath long-period prediction network to obtain channel multipath component prediction information of a plurality of future time instants of the terminal to be predicted; wherein the channel multipath component prediction information comprises channel multipath component prediction geometric information and channel multipath component prediction field intensity information.

[0010] Optionally, the method, wherein inputting the channel multipath component geometric information and the channel multipath component field intensity information of a plurality of historical time instants of the terminal to be predicted into a first feature extraction subnetwork of the multipath long-period prediction network to obtain a first feature map comprises:

[0011] obtaining the multipath long-period prediction network;

[0012] obtaining the channel multipath component geometric information and the channel multipath component field intensity information through a ray tracing simulation model or actual channel measurement;

[0013] inputting the channel multipath component geometric information and the channel multipath component field intensity information into the first feature extraction subnetwork of the multipath long-period prediction network for feature extraction, mining the relationship between the geometry and the field intensity of the channel multipath components, and obtaining the first feature map.

[0014] Optionally, the method, wherein the obtaining the multipath long-period prediction network comprises:

[0015] adopting a deep learning architecture to build a to-be-trained multipath long-period prediction network;

[0016] setting hyperparameters of the to-be-trained multipath long-period prediction network and determining a loss function;

[0017] obtaining channel multipath components of a plurality of continuous moving positions and constructing a time series data set of the channel multipath components;

[0018] Training the to-be-trained multipath long-period prediction network by using the loss function and the time series data set, to obtain the multipath long-period prediction network.

[0019] Optionally, the method, wherein, before inputting the coordinate information corresponding to the plurality of historical time instants of the to-be-predicted terminal and the scene geometric feature of the wireless propagation environment in which the to-be-predicted terminal is located into a second feature extraction subnetwork of the multipath long-period prediction network to obtain a second feature map, the method further comprises:

[0020] establishing an environment model of the wireless propagation environment in three dimensions;

[0021] obtaining scatterer information of the wireless propagation environment and motion information of the to-be-predicted terminal according to the environment model;

[0022] obtaining the scene geometric feature and the material electromagnetic parameter feature according to the scatterer information;

[0023] obtaining the coordinate information according to the motion information.

[0024] Optionally, the method, wherein the scatterer information comprises one or more of the following:

[0025] scatterer position;

[0026] scatterer shape;

[0027] scatterer size;

[0028] scatterer material electromagnetic parameter.

[0029] Optionally, the method, wherein inputting the first feature map, the second feature map and the third feature map into a sequence prediction subnetwork of the multipath long-period prediction network to obtain channel multipath component prediction information of the to-be-predicted terminal at a plurality of future time instants comprises:

[0030] inputting the first feature map, the second feature map and the third feature map into the sequence prediction subnetwork to obtain spliced features;

[0031] obtaining time series evolution rules by using a neural network model;

[0032] obtaining the channel multipath component prediction information at the plurality of future time instants according to the spliced features and the time series evolution rules.

[0033] Optionally, the method, wherein the channel multipath component prediction information comprises one or more of the following:

[0034] channel multipath component prediction geometric information;

[0035] channel multipath component predicted field strength information.

[0036] To achieve the above object, the embodiment of the present application provides an environment feature assisted channel multipath intelligent prediction device, which comprises:

[0037] The first acquisition module is configured to input channel multipath component geometric information and channel multipath component field strength information of a plurality of historical time points of a terminal to be predicted into a first feature extraction subnetwork of a multipath long-period prediction network to obtain a first feature map; wherein the first feature map comprises a geometric and field strength relationship of channel multipath components of the terminal to be predicted.

[0038] The second acquisition module is configured to input coordinate information corresponding to the plurality of historical time points of the terminal to be predicted and scene geometric features of a wireless propagation environment in which the terminal to be predicted is located into a second feature extraction subnetwork of the multipath long-period prediction network to obtain a second feature map; wherein the second feature map comprises a geometric and scene geometric feature relationship of the channel multipath components.

[0039] The third acquisition module is configured to input the coordinate information and material electromagnetic parameter features of the wireless propagation environment into a third feature extraction subnetwork of the multipath long-period prediction network to obtain a third feature map; wherein the third feature map comprises a field strength and scene material electromagnetic parameter feature relationship of the channel multipath components.

[0040] The fourth acquisition module is configured to input the first feature map, the second feature map and the third feature map into a sequence prediction subnetwork of the multipath long-period prediction network to obtain channel multipath component prediction information of a plurality of future time points of the terminal to be predicted; wherein the channel multipath component prediction information comprises channel multipath component predicted geometric information and channel multipath component predicted field strength information.

[0041] To achieve the above object, the embodiment of the present application provides an electronic device, which comprises a transceiver, a processor, a memory and a program or instruction stored on the memory and executable on the processor; wherein the processor implements the environment feature assisted channel multipath intelligent prediction method as described above when executing the program or instruction.

[0042] To achieve the above object, the embodiment of the present application provides a readable storage medium, which stores a program or instruction, wherein the program or instruction is executed by a processor to implement the steps in the environment feature assisted channel multipath intelligent prediction method as described above.

[0043] To achieve the above object, the embodiment of the present application provides a computer program product, which comprises computer instructions, and the computer instructions realize the steps of the environment feature assisted channel multipath intelligent prediction method when executed by a processor.

[0044] The beneficial effects of the above technical solutions of the present application are as follows:

[0045] In the embodiment of the present application, the channel multipath component geometric information, channel multipath component field strength information, coordinate information, scene geometric features and material electromagnetic parameter features of a plurality of historical time points of a terminal to be predicted are input into different feature extraction sub-networks of a multipath long-period prediction network, a plurality of feature maps including the relationships between the geometric and field strength of the channel multipath component of the terminal to be predicted, scene geometric features and scene material electromagnetic parameter features are obtained, and the plurality of feature maps are input into a sequence prediction sub-network of the multipath long-period prediction network to obtain channel multipath component prediction information at a plurality of future time points. The above embodiment builds a network for a channel multipath time series prediction task, and in terms of prediction capability, in order to further improve the prediction performance, environmental features including geometric and material features of the environment are introduced to mine the influence of the environment on electromagnetic wave propagation, and the prediction accuracy is further improved to ensure the long-period prediction capability. Thus, the problem that the existing technology lacks prediction of MPCs at future time points, including geometric information and field strength information of MPCs, in AI-based channel prediction is improved. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 A schematic diagram of the environment feature assisted channel multipath intelligent prediction method described in the embodiment of the present application;

[0047] Figure 2 A schematic diagram of the MPCs geometric and field strength variation law of the environment feature assisted channel multipath intelligent prediction method described in the embodiment of the present application;

[0048] Figure 3 A flowchart of the environment feature assisted channel multipath intelligent prediction method described in the embodiment of the present application;

[0049] Figure 4 A schematic diagram of the multipath long-period prediction network of the environment feature assisted channel multipath intelligent prediction method described in the embodiment of the present application;

[0050] Figure 5 One of the evaluation prediction result schematic diagrams of the environment feature assisted channel multipath intelligent prediction method described in the embodiment of the present application;

[0051] Figure 6 The second evaluation prediction result schematic diagram of the environment feature assisted channel multipath intelligent prediction method described in the embodiment of the present application;

[0052] Figure 7 This is a schematic diagram of the environmental feature-assisted intelligent multipath prediction device for channels according to an embodiment of the present invention. Detailed Implementation

[0053] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0054] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of the invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0055] In various embodiments of the present invention, it should be understood that the sequence number of each process described below does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0056] In addition, the terms "system" and "network" are often used interchangeably in this article.

[0057] In the embodiments provided in this application, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information.

[0058] For ease of understanding, the following describes some aspects of the embodiments of the present invention:

[0059] like Figure 1 As shown in the figure, an environmental feature-assisted intelligent prediction method for channel multipath propagation according to an embodiment of the present invention includes:

[0060] S10, input the geometric information of the channel multipath components and the field strength information of the channel multipath components of the terminal to be predicted at multiple historical moments into the first feature extraction subnetwork of the multipath long-period prediction network to obtain a first feature map; wherein, the first feature map includes the relationship between the geometry and field strength of the channel multipath components of the terminal to be predicted.

[0061] It should be noted that, as Figure 2 As shown in the example of electromagnetic wave reflection, the point of action of MPCs is defined as the interaction point between the electromagnetic wave and the building, such as the reflection points r1, r2, ..., r in the reflection example. nThe point of application of multipath propagation patterns (MPCs) contains complete geometric information about the MPCs, allowing for the reconstruction of the electromagnetic wave propagation path. The received field strength of the MPCs is the electric field strength received by the receiving antenna along that path. The point of application and received field strength of the MPCs vary depending on the vehicle's location, but they exhibit evolutionary patterns. Therefore, AI models such as Transformer and LSTM possess powerful sequential pattern learning capabilities, and these patterns are expected to be effectively learned by neural networks to predict future multipath propagation. Figure 3 As shown, in step S10, historical MPCs (i.e., channel multipath component geometric information and channel multipath component field strength information at multiple historical moments of the terminal to be predicted) are obtained and input into EAML-net subnetwork 1 (i.e., the first feature extraction subnetwork of the multipath long-period prediction network) to obtain high-dimensional feature F1 (i.e., the first feature map), thereby obtaining the first feature map that can reflect the relationship between the geometry and field strength of the channel multipath components. Figure 4 As shown, in sub-network 1: multipath feature extraction network, taking historical time k as an example, the coordinates G of the MPC action point at historical time k are... k (equivalent to the aforementioned reflection points r1, r2, ..., r) n (One of them), and the historical k-time MPC field strength E k The data are fed into a multilayer perceptron and then into a multi-head attention mechanism layer to obtain high-dimensional features F1 (i.e., the first feature map).

[0062] S20, the coordinate information of the terminal to be predicted at the multiple historical times and the scene geometric features of the wireless propagation environment in which the terminal to be predicted is located are input into the second feature extraction subnetwork of the multipath long-period prediction network to obtain a second feature map; wherein, the second feature map includes the relationship between the geometry of the channel multipath components and the scene geometric features;

[0063] It should be noted that, as Figure 2 As shown, the receiver moves continuously at different times, located at Rx1, ..., Rx i , ..., Rx n Multiple locations, the receiver being the terminal to be predicted. For example... Figure 4 As shown, in sub-network 2: the scene geometric feature extraction network, taking historical time k as an example, the scene geometric feature O and the historical time k Rx coordinate P are... k (equivalent to Rx1, ..., Rx of the receiver mentioned above) i , ..., Rx n One of the multiple locations is input into a multilayer perceptron and then into a multilayer convolutional block to obtain a geometric feature map F2 (i.e., the second feature map), thereby obtaining the second feature map that can reflect the relationship between the geometric features of the channel multipath components and the geometric features of the scene.

[0064] S30, input the coordinate information and the material electromagnetic parameter features of the wireless propagation environment into the third feature extraction subnetwork of the multipath long-period prediction network to obtain a third feature map; wherein, the third feature map includes the relationship between the field strength of the channel multipath component and the scene material electromagnetic parameter features.

[0065] It should be noted that, as Figure 3 As shown, in steps S20 and S30, scene geometry (i.e., the scene geometric features of the wireless propagation environment where the terminal to be predicted is located) and material electromagnetic parameter features (i.e., the material electromagnetic parameter features of the wireless propagation environment) are obtained and input into EAML-net subnetworks 2 and 3 (i.e., the second feature extraction subnetwork and the third feature extraction subnetwork of the multipath long-period prediction network), respectively, to obtain high-dimensional features F2 and F3 (i.e., the second feature map and the third feature map). Figure 4 As shown, in sub-network 3: scene material feature extraction network, taking historical time k as an example, the scene material electromagnetic parameter feature I and the historical time k Rx coordinate P are combined. k (equivalent to Rx1, ..., Rx of the receiver mentioned above) i , ..., Rx n One of the multiple locations is input into a multilayer perceptron and then into a multilayer convolutional block to obtain the material electromagnetic feature map F3 (i.e., the third feature map), thereby obtaining the third feature map that can reflect the relationship between the field strength of the channel multipath component and the electromagnetic parameter characteristics of the scene material.

[0066] S40, input the first feature map, the second feature map and the third feature map into the sequence prediction subnetwork of the multipath long-period prediction network to obtain channel multipath component prediction information for multiple future times of the terminal to be predicted; wherein, the channel multipath component prediction information includes channel multipath component prediction geometric information and channel multipath component prediction field strength information.

[0067] It should be noted that, as Figure 3 As shown, in step S40, F1, F2, and F3 (i.e., the first feature map, the second feature map, and the third feature map) are input into the EAML-net subnetwork 4 (i.e., the sequence prediction subnetwork of the multipath long-period prediction network) to predict multiple future MPCs (i.e., the channel multipath component prediction information of the terminal to be predicted). Figure 4As shown, in sub-network 4: multipath sequence prediction network, F1, F2, and F3 (i.e., the first feature map, the second feature map, and the third feature map) are concatenated and merged, input into the Transformer temporal prediction layer, and then input into a multi-layer convolutional block to obtain the channel multipath component prediction information of the terminal to be predicted at multiple future times. Taking future time t as an example, it is divided into the MPC action point coordinates G at future time t. t (i.e., the predicted geometric information of the channel multipath components) and the MPC field strength E at time t in the future. t (i.e., the predicted field strength information of the channel multipath components).

[0068] In this embodiment, the channel multipath component geometric information, channel multipath component field strength information, coordinate information, scene geometric features of the wireless propagation environment, and material electromagnetic parameter features of the terminal to be predicted at multiple historical moments are input into different subnetworks of the multipath long-period prediction network. Multiple feature maps, including the relationship between the geometry and field strength of the channel multipath components of the terminal to be predicted, scene geometric features, and scene material electromagnetic parameter features, are obtained respectively. These feature maps are then input into the multipath long-period prediction network to obtain channel multipath component prediction information for multiple future moments. The above embodiment constructs a network for time-series prediction tasks of channel multipath. Furthermore, to further improve prediction performance, environmental features, including the geometric and material features of the environment, are introduced to explore the influence of the environment on electromagnetic wave propagation, further improving prediction accuracy and ensuring long-period prediction capabilities. This addresses the problem in existing AI-based channel prediction technologies that lack predictions of future MPCs, including the geometric and field strength information of the MPCs.

[0069] Optionally, the method, wherein step S10 includes:

[0070] Obtain the multipath long-period prediction network;

[0071] The geometric information of the channel multipath component and the field strength information of the channel multipath component are obtained by ray tracing simulation model;

[0072] The geometric information and field strength information of the channel multipath components are input into the first feature extraction subnetwork of the multipath long-period prediction network for feature extraction, to explore the relationship between the geometry and field strength of the channel multipath components, and to obtain the first feature map.

[0073] In this embodiment, information such as the position, shape, size, and material of the scatterer in the wireless propagation environment is obtained. environment Antenna configuration C of transmitter and receiver (i.e., the terminal to be predicted) antennaThis includes the number of antennas, polarization, gain, and transmit power; the motion parameters C of the terminal to be predicted. move including velocity vector acceleration vector i = 1, 2, ..., m, where m is the number of moving objects, constructing a time-series dataset for MPCs. Ray tracing software such as WirelessInsite and BUPT-RT-SIM are used, based on C++. environment C antenna C move Ray tracing simulation was performed to obtain the MPCs geometric information G of the terminal to be predicted, taking historical time k as an example, from multiple historical time points of the prediction terminal. k (i.e., the geometric information of the channel multipath components) and field strength information E k (i.e., the channel multipath component field strength information). The historical MPCs geometric information G... k (i.e., the geometric information of the channel multipath components) and field strength information E k (i.e., the channel multipath component field strength information) is input into the multipath feature extraction module of sub-network 1 of LMPE-net (i.e., the first feature extraction sub-network of the multipath long-period prediction network) for automatic feature extraction, during which a 128-head attention mechanism is used to mine G. k and E k The relationship between the MPCs is determined, and finally the fused MPCs change pattern feature F1 (i.e., the first feature map) is input into sub-network 4 (i.e., the sequence prediction sub-network).

[0074] Optionally, the method, wherein obtaining the multipath long-period prediction network includes:

[0075] A deep learning architecture is used to build a multipath long-period prediction network to be trained.

[0076] Set the hyperparameters of the multipath long-period prediction network to be trained and determine the loss function;

[0077] Obtain the channel multipath components of multiple consecutive moving locations, and construct the time series dataset of the channel multipath components;

[0078] The multipath long-period prediction network is trained using the loss function and the time series dataset to obtain the multipath long-period prediction network.

[0079] In this embodiment, the LMPE-net network (i.e., the multipath long-period prediction network to be trained) is built using the PyTorch deep learning architecture. It includes sub-network 1 (multipath feature extraction module), sub-network 2 (scene geometry feature extraction module), sub-network 3 (scene material feature extraction module), and sub-network 4 (multipath sequence prediction module). The hyperparameters of LMPE-net are set, and the learning rate is set to 10. -4 The loss function is the sum of the mean square error of the geometric information G and the field strength information E of the MPCs at the prediction time:

[0080]

[0081] Among them, G m and E represents the true value and predicted value of the geometric information of MPCs, respectively. m and These represent the true and predicted values ​​of the MPCs field strength information, respectively. The time series dataset consists of multiple (more than 10,000) MPCs data points at consecutive moving positions. A 3D environment model is performed on the scatterers and environmental boundaries in the wireless propagation environment. Motion parameters of the terminal to be predicted in the scene are set, including information such as the position, shape, size, and material of the scatterers in the wireless propagation environment. environment Antenna configuration C of transmitter and receiver (i.e., the terminal to be predicted) antenna This includes the number of antennas, polarization, gain, and transmit power; the motion parameters C of the terminal to be predicted. move including velocity vector acceleration vector i = 1, 2, ..., m, where m is the number of moving objects. Ray tracing software such as WirelessInsite and BUPT-RT-SIM are used, based on C++. environment C antenna C move Perform ray tracing simulation to obtain the geometric information (Geo) of MPC at different times. i Field strength information Ele i Constructing a time series dataset Set for MPCs MPCs (i.e., the time series dataset). Geometric information is the coordinates R of the MPC's point of action. n , where n represents the order of action; the field strength information is the received field strength E of the MPC, that is: The multipath long-period prediction network to be trained is trained on a graphics card with more than 6GB of memory using the time series dataset, and the multipath long-period prediction network is obtained.

[0082] Optionally, the method further includes, prior to step S20:

[0083] Establish a three-dimensional environmental model of the wireless propagation environment;

[0084] Based on the environmental model, obtain the scatterer information of the wireless propagation environment and the motion information of the terminal to be predicted;

[0085] The scene geometric features and the material electromagnetic parameter features are obtained based on the scatterer information;

[0086] The coordinate information is obtained based on the motion information.

[0087] In this embodiment, based on C environment The following steps are taken to construct scene geometric features O (i.e., the scene geometric features) and material electrical parameter features I (i.e., the material electromagnetic parameter features). O is defined as tensor[L,W,2], where L and W represent the length and width of the scene in meters. O has two channels, representing the boundary and height features of the scene scatterers, respectively. I is also defined as tensor[L,W,2], where the L and W channels represent the dielectric constant and conductivity of the scene scatterer material, respectively. The coordinate information P of the terminal to be predicted (taking historical time k as an example) at time k is obtained. k P is defined as tensor[k,3,1], representing the three-dimensional coordinates (i.e., the coordinate information) of the terminal to be predicted at time k in history.

[0088] Optionally, in the method, the scatterer information includes one or more of the following:

[0089] Location of the scatterer;

[0090] The shape of the scatterer;

[0091] Scatterer size;

[0092] Electromagnetic parameters of the scatterer material.

[0093] In this embodiment, information such as the location, shape, size, and electromagnetic parameters of the scatterer in the wireless propagation environment is obtained and modeled to scale in modeling software such as SketchUp and Blender to obtain the scene's geometric features and the material's electromagnetic parameter features.

[0094] Optionally, the method, wherein step S40 includes:

[0095] The first feature map, the second feature map, and the third feature map are input into the sequence prediction sub-network to obtain spliced ​​features;

[0096] Using neural network models to obtain the evolution patterns of time series;

[0097] Based on the splicing characteristics and the time series evolution pattern, the channel multipath component prediction information for multiple future moments is obtained.

[0098] In this embodiment, the scene geometric feature O (i.e., the scene geometric feature) and the historical coordinates P are compared. k The coordinate information is input into the scene geometric feature extraction module of sub-network 2 of LMPE-net (i.e., the second feature extraction sub-network) for automatic feature extraction. During this process, four convolutional layers and one linear layer are used to extract geometric feature maps, resulting in geometric feature map F2 (i.e., the second feature map). The material electromagnetic parameter feature I (i.e., the material electromagnetic parameter feature) is compared with the historical coordinates P. k The coordinate information is input into the scene material feature extraction module of sub-network 3 of LMPE-net (i.e., the third feature extraction sub-network) for automatic feature extraction. During this process, four convolutional layers and one linear layer are used to extract the material electromagnetic feature map, resulting in material electromagnetic feature map F3 (i.e., the third feature map). Based on the MPCs change pattern feature F1 (i.e., the first feature map), geometric feature map F2 (i.e., the second feature map), and material electromagnetic feature map F3 (i.e., the third feature map), a neural network model such as Transformer is used to mine the time change pattern, obtain the time series evolution pattern, and predict the MPCs (i.e., the channel multipath component prediction information) at multiple future times (taking time t as an example). Specifically, this includes: 1) Inputting F1, F2, and F3 into the multipath sequence prediction module of sub-network 4 of LMPE-net (i.e., the sequence prediction sub-network) for feature concatenation to obtain the concatenated features. 2) Based on the spliced ​​features (i.e., the spliced ​​features), a neural network model, such as a two-layer codec architecture transformer, is used to mine the time series evolution patterns and predict the MPCs (i.e., the channel multipath component prediction information) at multiple future times (taking time t as an example), including G. t (i.e., the channel multipath component prediction geometric information) and E t (i.e., the predicted field strength information of the channel multipath component), to complete the entire process of predicting the time series of the channel multipath component.

[0099] Optionally, in the method, the channel multipath component prediction information includes one or more of the following:

[0100] Geometric information for channel multipath component prediction;

[0101] Predicting field strength information from channel multipath components.

[0102] In this embodiment, the MPCs (i.e., the channel multipath component prediction information) at future time t include G t(i.e., the channel multipath component prediction geometric information) and E t (i.e., the predicted field strength information of the channel multipath components).

[0103] It should be noted that the specific steps for evaluating the environmental feature-assisted intelligent channel multipath prediction method described in this embodiment of the invention are as follows:

[0104] 1) Obtain the propagation physical environment of a real street scene, such as the open-source OpenStreetMap website. The scene dimensions are 340 meters long, 280 meters wide, and 40 meters high. Define the transmitter as a stationary base station and the receiver as a moving object, both moving continuously within the scene at a speed of 36 km / h.

[0105] 2) The coordinate features P of the UE were constructed based on the Rx positions at 9000 time points. Then, the MPC information at 9000 time points was obtained through ray tracing simulation, and an MPCs dataset was constructed. Based on the 3D model, the geometric information O and the scene material features I were constructed.

[0106] 3) The dataset was divided into training, testing, and validation sets in a 7:1:2 ratio. The training set contained 6300 samples, the testing set contained 900 samples, and the validation set contained 1800 samples. Next, the network was trained on a suitable platform with a learning rate of 0.0008 and 200 epochs.

[0107] 4) Using a network based solely on the traditional transformer as the baseline, i.e., only sub-network 4, we also trained the performance of networks with only the multipath feature extraction module plus the multipath sequence prediction module (i.e., sub-networks 1 and 4), and networks with only the multipath feature extraction module plus the multipath sequence prediction module plus the scene geometric feature extraction module (i.e., sub-networks 1, 2, and 4) to demonstrate the impact of each module on prediction performance. Normalized root mean square error (NMSE) was used as the metric. Figure 5 and Figure 6 As shown.

[0108] from Figure 5 and Figure 6 As can be seen from the above, the environmental feature-assisted channel multipath intelligent prediction method in this embodiment of the invention reduces the average error of geometry and field strength at 10 time points by 31.4% and 40.5% respectively compared with the baseline, and proves that introducing scene information can significantly improve prediction accuracy.

[0109] Compared with the prior art, the embodiments of the present invention have the following advantages:

[0110] 1. Considering the need for accurate prediction of propagation multipath information in DTC, this paper fully utilizes the continuity of electromagnetic wave propagation in space and takes full advantage of the variation of the point of action and field strength of MPCs at different times. Based on an AI model, an intelligent prediction method for channel multipath assisted by environment features is proposed. The proposed Long-Term Multipath Prediction Network assisted by Environment (LMPE-net) solves the shortcomings of traditional algorithms that cannot predict MPCs at future times.

[0111] 2. By making full use of environmental information, such as the position and size of scatterers and the influence of material electromagnetic parameters on electromagnetic wave propagation, a scene geometric feature extraction module and a scene material feature extraction module were designed, which greatly improved the accuracy of MPCs prediction and ensured prediction performance over long periods (such as historical time: future time = 1:1).

[0112] 3. Compared to existing schemes, this study reveals the changes that the movement of objects in dynamic environments brings to MPCs, thereby affecting the characteristics of wireless channels. A fast, accurate, and physically interpretable channel prediction scheme is established.

[0113] like Figure 7 As shown, to achieve the above objectives, embodiments of the present invention provide an environment-feature-assisted intelligent channel multipath prediction device, comprising:

[0114] The first acquisition module 701 is used to input the geometric information of the channel multipath components and the field strength information of the channel multipath components of the terminal to be predicted at multiple historical moments into the first feature extraction subnetwork of the multipath long-period prediction network to obtain a first feature map; wherein, the first feature map includes the relationship between the geometry and field strength of the channel multipath components of the terminal to be predicted.

[0115] The second acquisition module 702 is used to input the coordinate information of the terminal to be predicted at the multiple historical times and the scene geometric features of the wireless propagation environment in which the terminal to be predicted is located into the second feature extraction sub-network of the multipath long-period prediction network to obtain a second feature map; wherein, the second feature map includes the relationship between the geometry of the channel multipath component and the scene geometric features.

[0116] The third acquisition module 703 is used to input the coordinate information and the material electromagnetic parameter features of the wireless propagation environment into the third feature extraction subnetwork of the multipath long-period prediction network to obtain a third feature map; wherein, the third feature map includes the relationship between the field strength of the channel multipath component and the scene material electromagnetic parameter features.

[0117] The fourth acquisition module 704 is used to input the first feature map, the second feature map and the third feature map into the sequence prediction subnetwork of the multipath long-period prediction network to acquire channel multipath component prediction information for multiple future times of the terminal to be predicted; wherein, the channel multipath component prediction information includes channel multipath component prediction geometric information and channel multipath component prediction field strength information.

[0118] Optionally, in the aforementioned apparatus, the first acquisition module 701 includes:

[0119] The first acquisition unit is used to acquire the multipath long-period prediction network;

[0120] The second acquisition unit is used to acquire the geometric information of the channel multipath component and the field strength information of the channel multipath component through a ray tracing simulation model.

[0121] The third acquisition unit is used to input the geometric information of the channel multipath component and the field strength information of the channel multipath component into the first feature extraction subnetwork of the multipath long-period prediction network for feature extraction, to explore the relationship between the geometry and field strength of the channel multipath component, and to obtain the first feature map.

[0122] Optionally, in the apparatus, the first acquiring unit includes:

[0123] The first processing component is used to build a multipath long-period prediction network to be trained using a deep learning architecture.

[0124] The first determining component is used to set the hyperparameters of the multipath long-period prediction network to be trained and to determine the loss function;

[0125] The second processing component is used to acquire channel multipath components from multiple consecutive moving locations and construct a time series dataset of the channel multipath components.

[0126] The first acquisition component is used to train the multipath long-period prediction network to be trained using the loss function and the time series dataset, and to acquire the multipath long-period prediction network.

[0127] Optionally, the device further includes:

[0128] The first processing module is used to establish a three-dimensional environmental model of the wireless propagation environment.

[0129] The fifth acquisition module is used to acquire the scatterer information of the wireless propagation environment and the motion information of the terminal to be predicted based on the environment model.

[0130] The sixth acquisition module is used to acquire the scene geometric features and the material electromagnetic parameter features based on the scatterer information;

[0131] The seventh acquisition module is used to acquire the coordinate information based on the motion information.

[0132] Optionally, in the aforementioned apparatus, the scatterer information includes one or more of the following:

[0133] Location of the scatterer;

[0134] The shape of the scatterer;

[0135] Scatterer size;

[0136] Electromagnetic parameters of the scatterer material.

[0137] Optionally, in the aforementioned apparatus, the fourth acquisition module 704 includes:

[0138] The fourth acquisition unit is used to input the first feature map, the second feature map and the third feature map into the sequence prediction sub-network to acquire spliced ​​features;

[0139] The fifth acquisition unit is used to obtain the evolution pattern of time series using a neural network model;

[0140] The sixth acquisition unit is used to acquire the channel multipath component prediction information for multiple future times based on the splicing features and the time series evolution law.

[0141] Optionally, in the aforementioned apparatus, the channel multipath component prediction information includes one or more of the following:

[0142] Geometric information for channel multipath component prediction;

[0143] Predicting field strength information from channel multipath components.

[0144] It should be noted that the apparatus provided in this embodiment of the invention can implement all the method steps implemented in the above method embodiment and can achieve the same technical effect. Therefore, the parts and beneficial effects that are the same as those in the method embodiment will not be described in detail here.

[0145] To achieve the above objectives, embodiments of the present invention provide an electronic device, including: a transceiver, a processor, a memory, and a program or instructions stored in the memory and executable on the processor; wherein, when the processor executes the program or instructions, it implements the environmental feature-assisted channel multipath intelligent prediction method as described above.

[0146] To achieve the above objectives, embodiments of the present invention provide a readable storage medium storing a program or instructions thereon, wherein the program or instructions, when executed by a processor, implement the steps in the environment feature-assisted channel multipath intelligent prediction method as described above.

[0147] To achieve the above objectives, embodiments of the present invention provide a computer program product, comprising computer instructions that, when executed by a processor, implement the steps of the environment feature-assisted intelligent prediction method for channel multipath as described above.

[0148] It should be further noted that the terminals described in this specification include, but are not limited to, smartphones, tablets, etc., and many of the functional components described are referred to as modules in order to emphasize the independence of their implementation.

[0149] In this embodiment of the invention, the module can be implemented in software so that it can be executed by various types of processors. For example, an identified executable code module may include one or more physical or logical blocks of computer instructions, which may be constructed as objects, procedures, or functions. Nevertheless, the executable code of the identified module does not need to be physically located together, but may include different instructions stored in different bits, which, when logically combined, constitute the module and achieve the module's intended purpose.

[0150] In practice, an executable code module can be a single instruction or many instructions, and can even be distributed across multiple different code segments, different programs, and across multiple memory devices. Similarly, operational data can be identified within the module and can be implemented in any suitable form and organized within any suitable type of data structure. This operational data can be collected as a single dataset or distributed across different locations (including different storage devices), and can exist, at least in part, solely as electronic signals within the system or network.

[0151] When a module can be implemented using software, considering the current level of hardware technology, modules that can be implemented in software can be implemented using hardware circuits by those skilled in the art to achieve the corresponding functions, without considering cost. These hardware circuits include conventional very-large-scale integrated circuits (VLSI) or gate arrays, as well as existing semiconductors such as logic chips and transistors, or other discrete components. Modules can also be implemented using programmable hardware devices, such as field-programmable gate arrays, programmable array logic, and programmable logic devices.

[0152] The exemplary embodiments described above are with reference to the accompanying drawings. Many different forms and embodiments are feasible without departing from the spirit and teachings of the invention. Therefore, the invention should not be construed as limiting the exemplary embodiments set forth herein. Rather, these exemplary embodiments are provided to make the invention complete and convey the scope of the invention to those skilled in the art. In these drawings, component dimensions and relative dimensions may be exaggerated for clarity. The terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. As used herein, unless clearly indicated otherwise, the singular forms “a,” “an,” and “the” are intended to include all such forms. It will be further understood that the terms “comprising” and / or “including”, when used in this specification, indicate the presence of the stated features, integers, steps, operations, components, and / or elements, but do not exclude the presence or addition of one or more other features, integers, steps, operations, components, and / or groups thereof. Unless otherwise indicated, when stated, a range of values ​​includes the upper and lower limits of the range and any subranges in between.

[0153] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A channel multipath intelligent prediction method assisted by environmental features, characterized in that, include: The geometric information of the channel multipath components and the field strength information of the channel multipath components at multiple historical moments of the terminal to be predicted are input into the first feature extraction subnetwork of the multipath long-period prediction network to obtain a first feature map; wherein, the first feature map includes the relationship between the geometry and field strength of the channel multipath components of the terminal to be predicted. The coordinate information of the terminal to be predicted at the multiple historical moments and the scene geometric features of the wireless propagation environment in which the terminal to be predicted is located are input into the second feature extraction subnetwork of the multipath long-period prediction network to obtain a second feature map; wherein, the second feature map includes the relationship between the geometry of the channel multipath components and the scene geometric features. The coordinate information and the material electromagnetic parameter features of the wireless propagation environment are input into the third feature extraction subnetwork of the multipath long-period prediction network to obtain a third feature map; wherein, the third feature map includes the relationship between the field strength of the channel multipath component and the scene material electromagnetic parameter features. The first feature map, the second feature map, and the third feature map are input into the sequence prediction subnetwork of the multipath long-period prediction network to obtain channel multipath component prediction information for multiple future times of the terminal to be predicted; wherein, the channel multipath component prediction information includes channel multipath component prediction geometric information and channel multipath component prediction field strength information.

2. The method according to claim 1, characterized in that, The geometric information of the channel multipath components and the field strength information of the channel multipath components at multiple historical moments of the terminal to be predicted are input into the first feature extraction subnetwork of the multipath long-period prediction network to obtain the first feature map, including: Obtain the multipath long-period prediction network; The geometric information of the channel multipath component and the field strength information of the channel multipath component are obtained by ray tracing simulation model; The geometric information and field strength information of the channel multipath components are input into the first feature extraction subnetwork of the multipath long-period prediction network for feature extraction, to explore the relationship between the geometry and field strength of the channel multipath components, and to obtain the first feature map.

3. The method according to claim 2, characterized in that, The process of obtaining the multipath long-period prediction network includes: A deep learning architecture is used to build a multipath long-period prediction network to be trained. Set the hyperparameters of the multipath long-period prediction network to be trained and determine the loss function; Obtain the channel multipath components of multiple consecutive moving locations, and construct the time series dataset of the channel multipath components; The multipath long-period prediction network is trained using the loss function and the time series dataset to obtain the multipath long-period prediction network.

4. The method according to claim 1, characterized in that, Before inputting the coordinate information of the terminal to be predicted at the multiple historical times and the scene geometric features of the wireless propagation environment in which the terminal to be predicted is located into the second feature extraction subnetwork of the multipath long-period prediction network to obtain the second feature map, the method further includes: Establish a three-dimensional environmental model of the wireless propagation environment; Based on the environmental model, obtain the scatterer information of the wireless propagation environment and the motion information of the terminal to be predicted; The scene geometric features and the material electromagnetic parameter features are obtained based on the scatterer information; The coordinate information is obtained based on the motion information.

5. The method according to claim 4, characterized in that, The scatterer information includes one or more of the following: Location of the scatterer; The shape of the scatterer; Scatterer size; Electromagnetic parameters of the scatterer material.

6. The method according to claim 1, characterized in that, The first feature map, the second feature map, and the third feature map are input into the sequence prediction subnetwork of the multipath long-period prediction network to obtain channel multipath component prediction information for multiple future time points of the terminal to be predicted, including: The first feature map, the second feature map, and the third feature map are input into the sequence prediction sub-network to obtain spliced ​​features; Using neural network models to obtain the evolution patterns of time series; Based on the splicing characteristics and the time series evolution pattern, the channel multipath component prediction information for multiple future moments is obtained.

7. The method according to claim 6, characterized in that, The channel multipath component prediction information includes one or more of the following: Geometric information for channel multipath component prediction; Predicting field strength information from channel multipath components.

8. An environmentally assisted intelligent multipath prediction device for channels, characterized in that, include: The first acquisition module is used to input the geometric information of the channel multipath components and the field strength information of the channel multipath components of the terminal to be predicted at multiple historical moments into the first feature extraction subnetwork of the multipath long-period prediction network to obtain a first feature map; wherein, the first feature map includes the relationship between the geometry and field strength of the channel multipath components of the terminal to be predicted. The second acquisition module is used to input the coordinate information of the terminal to be predicted at the multiple historical times and the scene geometric features of the wireless propagation environment in which the terminal to be predicted is located into the second feature extraction subnetwork of the multipath long-period prediction network to obtain a second feature map; wherein, the second feature map includes the relationship between the geometry of the channel multipath component and the scene geometric features. The third acquisition module is used to input the coordinate information and the material electromagnetic parameter features of the wireless propagation environment into the third feature extraction subnetwork of the multipath long-period prediction network to obtain a third feature map; wherein, the third feature map includes the relationship between the field strength of the channel multipath component and the scene material electromagnetic parameter features. The fourth acquisition module is used to input the first feature map, the second feature map, and the third feature map into the sequence prediction subnetwork of the multipath long-period prediction network to acquire channel multipath component prediction information for multiple future times of the terminal to be predicted; wherein, the channel multipath component prediction information includes channel multipath component prediction geometric information and channel multipath component prediction field strength information.

9. An electronic device, comprising: A transceiver, a processor, a memory, and a program or instructions stored in the memory and executable on the processor; characterized in that, when the processor executes the program or instructions, it implements the environmental feature-assisted channel multipath intelligent prediction method as described in any one of claims 1-7.

10. A readable storage medium having a program or instructions stored thereon, characterized in that, When the program or instructions are executed by the processor, they implement the steps in the environmental feature-assisted intelligent prediction method for channel multipath as described in any one of claims 1-7.

11. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the steps of the environmental feature-assisted intelligent prediction method for channel multipath as described in any one of claims 1-7.

Images