Channel scene classification method based on multi-modal scene features and application
By constructing a multimodal scene feature dataset and performing dimensionality reduction, combined with clustering algorithms, the subjectivity and adaptability issues of scene classification in traditional channel modeling are solved, achieving adaptive classification and accurate modeling of channel scenes.
Patent Information
- Application Number
- CN202510456119.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-11
- Publication Date
- 2026-06-16
AI Technical Summary
In traditional channel modeling methods, scene classification based on geometric and physical environmental characteristics is subjective, which may lead to different classification results for the same geographical area due to differences in evaluation criteria. Furthermore, data-driven methods cannot form a clear correspondence with specific physical scenes, affecting the accuracy and adaptability of channel models.
By acquiring channel impulse response data and 3D topographic maps, channel characteristics and geometric and physical environment feature parameters are extracted to construct a multimodal scene feature dataset. Principal component analysis is used for dimensionality reduction, and clustering algorithms are combined to classify channel scenes, thereby achieving adaptive classification of channel scenes.
It improves the accuracy and adaptability of the channel model, enabling accurate and objective classification of channel scenarios, reducing adaptability issues during engineering deployment, and enhancing the accuracy of the channel model.
Smart Images
Figure CN122220920A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wireless channel modeling, and in particular to a channel scene classification method and its application based on multimodal scene features. Background Technology
[0002] In wireless communication systems, the wireless channel, as a core uncontrollable factor affecting communication quality, directly influences system performance. In practical engineering, channel models for specific scenarios are typically constructed to characterize the propagation patterns of electromagnetic waves, thereby guiding the design and optimization of communication systems. To achieve accurate modeling, channel measurement experiments are conducted to obtain measured data, and scenario-based channel models are established based on the measured results of key parameters such as multipath fading and path loss. As a key tool in communication system development, wireless channel modeling has multiple values: First, by predicting characteristics such as delay spread and Doppler shift in signal propagation, it provides theoretical support for network deployment and parameter optimization; second, at the standards development level, the modeling results provide a scientific basis for communication protocol design; furthermore, this technology also supports the development and verification of new communication algorithms, equipment performance testing, and the evaluation of cutting-edge technologies such as large-scale MIMO. It can be said that accurate channel modeling is not only the technological cornerstone for improving communication quality but also a crucial driving force for the iteration of wireless communication technologies.
[0003] Traditional channel modeling methods typically require channel measurements to be performed for specific scenarios, with scenario classification primarily based on differences in the geometric and physical environment. For example, the International Telecommunication Union Radiocommunication Sector (ITU-R) classifies cities into four categories—suburbs, ordinary cities, densely populated cities, and high-rise cities—based on building height and distribution characteristics. 3GPP, in its TR38.901 standard, further subdivides indoor scenarios into typical environments such as offices, gyms, stadiums, and factories. While this method of constructing a scenario classification system based on geometric and physical environmental characteristics facilitates subsequent model calls and standardized applications, it suffers from significant subjectivity: the same geographical area may produce different classification results due to differences in evaluation standards. Furthermore, when multiple scenarios with different channel characteristics are forcibly merged into a single category, it directly restricts the accuracy of the final channel model. To overcome the limitations of traditional classification methods, researchers have begun exploring new data-driven approaches. By deeply analyzing channel characteristic parameters, machine learning algorithms are used to achieve intelligent identification and classification of channel scenarios. However, such methods essentially only classify based on differences in channel characteristics and cannot establish a clear correspondence with specific physical scenarios, leading to challenges in scenario adaptability during engineering deployment of channel models. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this invention provides a channel scene classification method and its application based on multimodal scene features.
[0005] To achieve the above technical objectives, the specific technical solution adopted by the present invention is as follows: On the one hand, the present invention provides a channel scene classification method based on multimodal scene features, comprising the following steps: Acquire channel impulse response data and 3D terrain maps; Determine the steady-state interval of the channel based on channel impulse response data; Based on the steady-state interval of the channel, characteristic parameters of the channel are extracted, including root mean square delay spread, root mean square Doppler spread and Rice K factor; Geometric and physical environment characteristic parameters are calculated based on three-dimensional topographic maps, including average building height, building density, average vegetation height, vegetation density, and street width. By combining channel characteristic parameters and geometric and physical environment characteristic parameters, a multimodal scene feature dataset is constructed; Principal component analysis was used to reduce the dimensionality of a multimodal scene feature dataset. Clustering algorithms are used to perform clustering analysis on the dimensionality-reduced multimodal scene feature dataset, with each category corresponding to a channel scene; The channel scene obtained by clustering is parameterized by using the geometric and physical environment feature parameters in the multimodal scene feature dataset.
[0006] Furthermore, the steady-state interval of the channel is determined according to the following steps: Calculate the correlation coefficient of the power delay spectra of two channel snapshots based on channel impulse response data:
[0007] in, , for Channel impulse response at the location; N The average window length for calculating power; The maximum time interval between two channel snapshot power delay spectra reaching a preset threshold is used as the steady-state interval of the channel.
[0008] Furthermore, the root mean square delay spread is calculated according to the following formula:
[0009] in, For the first i The time delay of a multipath; For the first i Power of multi-path strips.
[0010] Furthermore, the root-mean-square Doppler extension is calculated according to the following formula:
[0011] in, The first Doppler power spectrum i Doppler frequency shift at each frequency shift point; The first Doppler power spectrum i The power at each frequency shift point.
[0012] Furthermore, the Rice K-factor is calculated according to the following formula:
[0013] in, It is a collection of multipath power.
[0014] Furthermore, the multimodal scene feature dataset can be represented as:
[0015] in, This represents the average building height. Building density; This represents the average vegetation height. Vegetation density; The width of the street.
[0016] Furthermore, the multimodal scene feature dataset is subjected to dimensionality reduction processing according to the following steps: Standardize the multimodal scene feature dataset:
[0017] in, For the first i 3D scene feature data, and The first i Mean and standard deviation of 3D scene feature data; Calculate the correlation coefficient matrix of the multimodal scene feature dataset:
[0018] in, The length of the dataset; Perform eigenvalue decomposition on the correlation coefficient matrix to obtain eigenvalues and eigenvectors:
[0019] in, For the first i One eigenvalue; For the first i The eigenvectors corresponding to each eigenvalue; Based on eigenvalue calculation, the first i Contribution rate of each principal component
[0020]
[0021] The top performers were selected based on their cumulative contribution rate (≥95%). One principal component, and utilize the previous The eigenvectors constitute the projection matrix. W Based on projection matrix W Obtain the dimensionality-reduced multimodal scene feature dataset Y
[0022] in, P The dimension of the multimodal scene feature dataset after dimensionality reduction.
[0023] Furthermore, the clustering algorithms include K-means algorithm, GMM algorithm, and ISODATA algorithm.
[0024] Furthermore, cluster analysis of the dimensionality-reduced multimodal scene feature dataset can be represented as follows:
[0025] in, For the first m Category labels corresponding to group scene feature data; This is the set of category labels corresponding to each set of scene feature data.
[0026] On the other hand, the present invention also provides an application of a channel scene classification method based on multimodal scene features, which is used to classify urban scenes.
[0027] Compared with the prior art, the beneficial technical effects of this invention are as follows: The present invention provides a channel scene classification method and application based on multimodal scene features. In the process of channel scene classification, both geometric and physical environment features and channel characteristic features are considered simultaneously. A multimodal scene feature dataset is established based on both. At the same time, the multimodal scene feature dataset is subjected to dimensionality reduction processing to reduce the impact of high-dimensional data on channel scene classification. Then, the geometric and physical environment feature parameters in the dimensionality-reduced multimodal scene feature dataset are used to parameterize the clustered channel scenes, thereby realizing adaptive channel scene classification in complex environments.
[0028] Traditional channel scene classification schemes based on geometric and physical environmental characteristics are overly subjective. Different evaluation criteria may lead to different classification results for the same geographical area. Furthermore, forcing multiple scenes with different channel characteristics into a single category directly limits the accuracy of the final channel model. In contrast, this invention is objective, accurately and objectively classifying channel scenes and effectively improving the accuracy of the channel model.
[0029] Existing data-driven scenario classification schemes generally only consider channel characteristics, and the classification results cannot form a clear correspondence with specific physical scenarios, leading to adaptability issues in channel model deployment during engineering. This invention, however, considers both geometric and physical environment features and channel characteristics, effectively solving the adaptability problem of existing channel models in engineering deployment and enabling accurate channel scenario classification. Attached Figure Description
[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.
[0031] Figure 1 A flowchart of a channel scene classification method based on multimodal scene features provided in one embodiment; Figure 2 A three-dimensional topographic map provided as an embodiment; Figure 3 This is a diagram showing the clustering analysis results of a dimensionality-reduced multimodal scene feature dataset provided in one embodiment. Detailed Implementation
[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0033] Reference Figure 1 One embodiment provides a channel scene classification method based on multimodal scene features, including the following steps: Acquire channel impulse response data and 3D terrain maps; Determine the steady-state interval of the channel based on channel impulse response data; Based on the steady-state interval of the channel, characteristic parameters of the channel are extracted, including root mean square delay spread, root mean square Doppler spread and Rice K factor; Geometric and physical environment characteristic parameters are calculated based on three-dimensional topographic maps, including average building height, building density, average vegetation height, vegetation density, and street width. By combining channel characteristic parameters and geometric and physical environment characteristic parameters, a multimodal scene feature dataset is constructed; Principal component analysis was used to reduce the dimensionality of a multimodal scene feature dataset. Clustering algorithms are used to perform clustering analysis on the dimensionality-reduced multimodal scene feature dataset, with each category corresponding to a channel scene; The channel scene obtained by clustering is parameterized by using the geometric and physical environment feature parameters in the multimodal scene feature dataset.
[0034] Channel impulse response data is obtained through channel measurements. Based on the channel impulse response data, the power delay spectra of two channel snapshots (i.e., and Correlation coefficient:
[0035] in, , for Channel impulse response at the location; N The average window length for calculating power; The maximum time interval between two channel snapshot power delay spectra reaching a preset threshold is used as the steady-state interval of the channel.
[0036] Based on the steady-state interval of the channel, channel characteristic parameters are extracted, including root mean square delay spread, root mean square Doppler spread, and Rice K-factor; wherein, the root mean square delay spread is calculated according to the following formula:
[0037] in, For the first i Delay of multiple paths; For the first i Power of multi-path strips; The root-mean-square Doppler extension is calculated according to the following formula:
[0038] in, The first Doppler power spectrum i Doppler frequency shift at each frequency shift point; The first Doppler power spectrum i Power at each frequency shift point; The Rice K-factor is calculated according to the following formula:
[0039] in, It is a collection of multipath power.
[0040] Reference Figure 2 This is a three-dimensional topographic map of an urban scene provided in one embodiment. Based on the three-dimensional topographic map, geometric and physical environment characteristic parameters are calculated, including average building height, building density, average vegetation height, vegetation density, and street width.
[0041] Average building height Building density is defined as the average height of all buildings within a certain range; Defined as the ratio of the building footprint to the total area within a certain range; average vegetation height. Defined as the average height of all vegetation within a certain range; vegetation density. Defined as the ratio of the area occupied by vegetation on the street to the total area; street width Defined as the distance between buildings on both sides of a road.
[0042] By combining channel characteristic parameters and geometric-physical environment characteristic parameters, a multimodal scene feature dataset is constructed, which can be represented as:
[0043] in, This represents the average building height. Building density; This represents the average vegetation height. Vegetation density; The width of the street.
[0044] By considering both geometric and physical environment features and channel characteristics during the channel scenario classification process, a multimodal scenario feature dataset is established. This dataset objectively and effectively integrates geometric and physical environment features and channel characteristics, making the channel scenario classification more closely reflect the actual situation and reducing adaptability issues during channel model engineering deployment.
[0045] In one embodiment, the multimodal scene feature dataset is subjected to dimensionality reduction processing according to the following steps: Standardize the multimodal scene feature dataset:
[0046] in, For the first i 3D scene feature data, and The firsti Mean and standard deviation of 3D scene feature data; Calculate the correlation coefficient matrix of the multimodal scene feature dataset:
[0047] in, The length of the dataset; Perform eigenvalue decomposition on the correlation coefficient matrix to obtain eigenvalues and eigenvectors:
[0048] in, For the first i One eigenvalue; For the first i The eigenvectors corresponding to each eigenvalue; Based on eigenvalue calculation, the first i Contribution rate of each principal component
[0049]
[0050] The top performers were selected based on their cumulative contribution rate (≥95%). One principal component, and utilize the previous The eigenvectors constitute the projection matrix. W Based on projection matrix W Obtain the dimensionality-reduced multimodal scene feature dataset Y
[0051] in, P The dimension of the multimodal scene feature dataset after dimensionality reduction.
[0052] By reducing the dimensionality of the multimodal scene feature dataset, the impact of high-dimensional datasets on scene classification is reduced.
[0053] Clustering algorithms are used to perform cluster analysis on the dimensionality-reduced multimodal scene feature dataset; the clustering algorithms include K-means algorithm, GMM algorithm, and ISODATA algorithm.
[0054] This embodiment uses the K-means algorithm to perform cluster analysis on the dimensionality-reduced multimodal scene feature dataset. The cluster analysis process can be represented as follows:
[0055] in, For the first m Category labels corresponding to group scene feature data; This is the set of category labels corresponding to each set of scene feature data.
[0056] Reference Figure 3 The figure shows the clustering analysis results of the multimodal scene feature dataset after dimensionality reduction. As can be seen from the figure, this invention uses both geometric and physical environment features and channel characteristics as inputs to the clustering algorithm. The output channel scenes not only have similar geometric and physical environment features, but also similar channel features. Therefore, this invention can achieve adaptive channel scene classification in complex environments.
[0057] Matters not covered in this invention are common knowledge.
[0058] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0059] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
[0060] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A channel scene classification method based on multimodal scene features, characterized in that, Includes the following steps: Acquire channel impulse response data and 3D terrain maps; Determine the steady-state interval of the channel based on channel impulse response data; Based on the steady-state interval of the channel, characteristic parameters of the channel are extracted, including root mean square delay spread, root mean square Doppler spread and Rice K factor; Geometric and physical environment characteristic parameters are calculated based on three-dimensional topographic maps, including average building height, building density, average vegetation height, vegetation density, and street width. By combining channel characteristic parameters and geometric and physical environment characteristic parameters, a multimodal scene feature dataset is constructed; Principal component analysis was used to reduce the dimensionality of a multimodal scene feature dataset. Clustering algorithms are used to perform clustering analysis on the dimensionality-reduced multimodal scene feature dataset, with each category corresponding to a channel scene; The channel scene obtained by clustering is parameterized by using the geometric and physical environment feature parameters in the multimodal scene feature dataset.
2. The channel scene classification method based on multimodal scene features as described in claim 1, characterized in that, The steady-state interval of the channel is determined according to the following steps: Calculate the correlation coefficient of the power delay spectra of two channel snapshots based on channel impulse response data: in, , for Channel impulse response at the location; N The average window length for calculating power; The maximum time interval between two channel snapshot power delay spectra reaching a preset threshold is used as the steady-state interval of the channel.
3. The channel scene classification method based on multimodal scene features as described in claim 1, characterized in that, The root mean square delay spread is calculated according to the following formula: in, For the first i Delay of multiple paths; For the first i Power of multi-path strips.
4. The channel scene classification method based on multimodal scene features as described in claim 1, characterized in that, The root-mean-square Doppler extension is calculated according to the following formula: in, The first Doppler power spectrum i Doppler frequency shift at each frequency shift point; The first Doppler power spectrum i The power at each frequency shift point.
5. The channel scene classification method based on multimodal scene features as described in claim 1, characterized in that, The Rice K-factor is calculated according to the following formula: in, It is a collection of multipath power.
6. The channel scene classification method based on multimodal scene features as described in claim 1, characterized in that, The multimodal scene feature dataset can be represented as: in, This represents the average building height. Building density; This represents the average vegetation height. Vegetation density; The width of the street.
7. The channel scene classification method based on multimodal scene features as described in claim 1, characterized in that, The multimodal scene feature dataset is dimensionality reduced according to the following steps: Standardize the multimodal scene feature dataset: in, For the first i 3D scene feature data, and The first i Mean and standard deviation of 3D scene feature data; Calculate the correlation coefficient matrix of the multimodal scene feature dataset: in, The length of the dataset; Perform eigenvalue decomposition on the correlation coefficient matrix to obtain eigenvalues and eigenvectors: in, For the first i One eigenvalue; For the first i The eigenvectors corresponding to each eigenvalue; Based on eigenvalue calculation, the first i Contribution rate of each principal component The top performers were selected based on their cumulative contribution rate (≥95%). One principal component, and utilize the previous The eigenvectors constitute the projection matrix. W Based on projection matrix W Obtain the dimensionality-reduced multimodal scene feature dataset Y in, P The dimension of the multimodal scene feature dataset after dimensionality reduction.
8. The channel scene classification method based on multimodal scene features as described in claim 1, characterized in that, The clustering algorithms include K-means algorithm, GMM algorithm, and ISODATA algorithm.
9. The channel scene classification method based on multimodal scene features as described in claim 1, characterized in that, Cluster analysis of the dimensionality-reduced multimodal scene feature dataset can be represented as: in, For the first m Category labels corresponding to group scene feature data; This is the set of category labels corresponding to each set of scene feature data.
10. The application of a channel scene classification method based on multimodal scene features, characterized in that, The urban scene is classified using the channel scene classification method based on multimodal scene features as described in any one of claims 1-9.