Intelligently aware wireless network dynamic coverage optimization system

Abstract

The application discloses a wireless network dynamic coverage optimization system with intelligent sensing, and belongs to the technical field of wireless networks.The system comprises a user management module, a coverage sensing module, an analysis and evaluation module, a control module and an adaptive module.The user equipment position information is mapped to a Riemann manifold, geometric features and topological features are extracted, and a manifold convolution neural network is constructed, so that intelligent sensing, accurate evaluation and dynamic optimization of network coverage are realized.The core innovation of the system lies in Riemann manifold wireless signal coverage space modeling based on differential geometry theory and a coverage quality deep learning evaluation model based on a manifold convolution network.The network coverage quality is comprehensively evaluated by calculating a coverage cavity factor, an overlapping coverage factor, a hot area coverage factor and a user perception factor, and dynamic optimization is realized through adaptive antenna parameter adjustment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless networks, particularly to a smart sensing wireless network dynamic coverage optimization system. BACKGROUND

[0002] With the rapid development of mobile communication and Internet of Things technology, wireless network coverage quality has become a key factor in user experience and service quality. Traditional wireless network coverage optimization methods mainly rely on manual experience and static models, which have the following shortcomings: first, traditional methods usually use Euclidean space distance attenuation model, which cannot accurately express the signal propagation characteristics in complex environments; second, there is a lack of precise evaluation ability for user actual perception, leading to a disconnection between optimization effect and user experience; finally, network parameter adjustment is mostly manual intervention, with slow response speed, making it difficult to adapt to rapidly changing network environment and user demand.

[0003] In the prior art, common network coverage optimization methods include optimization based on drive testing, optimization based on key performance indicators (KPI), and automatic parameter adjustment. These methods have improved network coverage quality to some extent, but still have obvious shortcomings in terms of perception ability, evaluation accuracy and adjustment flexibility. Especially in high-density user areas, complex building environments and rapidly changing user distribution scenarios, traditional methods are difficult to provide timely and effective coverage optimization.

[0004] Therefore, there is an urgent need for a system that can intelligently perceive network coverage status, accurately evaluate coverage quality and dynamically optimize coverage parameters to meet the demand for high-quality coverage of modern wireless networks. SUMMARY

[0005] The purpose of the present application is to provide a smart sensing wireless network dynamic coverage optimization system, aiming to overcome the shortcomings of the prior art and achieve intelligent perception, accurate evaluation and dynamic optimization of wireless network coverage.

[0006] The present application proposes a smart sensing wireless network dynamic coverage optimization system, comprising:

[0007] A user management module for receiving user equipment access requests and performing authentication, collecting user equipment location information, identifying the type of user equipment, and transmitting the collected user data to an analysis and evaluation module;

[0008] A coverage perception module in communication connection with the user management module for detecting user equipment data fluctuations, identifying access hotspots and uncovered areas, and transmitting the detection results to the analysis and evaluation module;

[0009] The analysis and evaluation module is in communication connection with the user management module and the coverage awareness module, is used for mapping the user equipment position information into a Riemann manifold, extracting geometric features and topological features of the Riemann manifold, constructing a manifold convolution neural network, calculating a coverage hole factor, an overlapping coverage factor, a hot area coverage factor and a user perception factor according to the geometric features and the topological features, generating a coverage evaluation result, and sending the coverage evaluation result to the control module;

[0010] The control module is in communication connection with the analysis and evaluation module, is used for receiving the coverage evaluation result, determining a control strategy based on the coverage evaluation result, generating a control instruction, and sending the control instruction to the adaptive module.

[0011] The adaptive module is in communication connection with the control module, is used for receiving the control instruction, adjusting an antenna position and direction according to the control instruction, dynamically adjusting a transmitting power and an antenna gain, and feeding back an adjustment result to the analysis and evaluation module.

[0012] Preferably, the analysis and evaluation module comprises:

[0013] A manifold construction unit is used for receiving the user equipment position information, pre-processing user position data, constructing an adjacency relation matrix, calculating a local metric tensor, and generating a discrete Riemann manifold.

[0014] A geometric feature extraction unit is in communication connection with the manifold construction unit, is used for calculating a Gaussian curvature and an average curvature on the discrete Riemann manifold, extracting a geodesic line feature, analyzing a topological structure of the discrete Riemann manifold, and generating a geometric feature vector and a topological feature vector.

[0015] A manifold convolution network unit is in communication connection with the geometric feature extraction unit, is used for receiving the geometric feature vector and the topological feature vector, extracting spatial features through a manifold convolution layer, performing dimension reduction processing through a manifold pooling layer, applying a manifold attention mechanism to identify a key area, and fusing multi-scale feature information.

[0016] A coverage evaluation decision unit is in communication connection with the manifold convolution network unit, is used for calculating a coverage hole factor, an overlapping coverage factor, a hot area coverage factor and a user perception factor based on the multi-scale feature information respectively, comprehensively evaluating network coverage quality, and generating the coverage evaluation result.

[0017] Preferably, the manifold construction unit comprises:

[0018] A data pre-processing module is used for filtering noise and removing outliers from the user equipment position information, converting geographic coordinates into standardized coordinates, and generating a standardized position point set.

[0019] An adjacency relationship construction module, in communication connection with the data preprocessing module, is configured to calculate a spatial relationship between points based on the standardized position point set, determine a neighborhood range, and construct an adjacency matrix.

[0020] A metric tensor calculation module, in communication connection with the adjacency relationship construction module, is configured to calculate a local metric tensor for each point in the standardized position point set, the local metric tensor reflecting anisotropic characteristics of signal propagation.

[0021] A manifold generation module, in communication connection with the metric tensor calculation module, is configured to combine the standardized position point set, the adjacency matrix, and the local metric tensor into a discrete Riemannian manifold, and construct a multi-scale manifold representation.

[0022] As a preference, the manifold convolution network unit comprises:

[0023] An input layer configured to receive the geometric feature vector and the topological feature vector, and organize the features into a data structure supporting manifold operations;

[0024] A manifold convolution layer, in communication connection with the input layer, is configured to define a convolution kernel based on a geodesic distance, adaptively adjust parameters of the convolution kernel according to local geometric characteristics, and extract local features;

[0025] A manifold pooling layer, in communication connection with the manifold convolution layer, is configured to implement adaptive pooling based on curvature, maintain high resolution for high-curvature areas, and down-sample low-curvature areas;

[0026] A manifold attention layer, in communication connection with the manifold pooling layer, is configured to calculate attention weights based on geometric features, and automatically focus on key areas of network coverage problems;

[0027] A multi-scale feature fusion layer, in communication connection with the manifold attention layer, is configured to construct a manifold feature pyramid, fuse local and global coverage information of different scales, and generate a comprehensive feature representation.

[0028] As a preference, the coverage evaluation decision unit calculates a coverage factor by the following way:

[0029] A hole factor , wherein represents the number of devices in the user equipment successfully connected to the network, represents the total number of user equipment;

[0030] An overlap factor , wherein A represents the number of user equipment with overlapping coverage.

[0031] A hot zone factor , wherein represents the number of user equipments covered by the hotspot, A represents the total number of user equipments covered by the hotspot;

[0032] User perception factor wherein represents the average throughput of the user equipment, T represents the measurement time, and V represents the average speed of the user equipment in the area;

[0033] When F1<0.5, F2≥0.5, F3<0.5, or F4<0.5, it is determined that the network coverage requirement is not met.

[0034] Preferably, the user management module comprises:

[0035] A user access authentication unit configured to authenticate the user equipment by using a username / password through a Basic authentication mode in an HTTP protocol;

[0036] A user location processing unit communicatively connected to the user access authentication unit and configured to obtain geographical location information of the user equipment by using a GPS positioning technology, calibrate the location information by using LBS information provided by a cellular mobile network or third-party map platform, and control the positioning error to be not greater than 500 meters;

[0037] A user classification analysis unit communicatively connected to the user location processing unit and configured to divide the user equipment into a normal terminal type, a multimedia type, and a VR type, wherein the normal terminal type user equipment uses a data volume as a basis for judgment, and the multimedia type and the VR type user equipment use a traffic size and a throughput as a basis for distinction.

[0038] Preferably, the coverage perception module comprises:

[0039] A device detection unit configured to detect connection and online status of the user equipment to the access hotspot by periodically sending an ICMP echo request and receiving an ICMP echo response, and determine whether the network coverage is blocked according to the online status of the user equipment;

[0040] A hole identification unit communicatively connected to the device detection unit and configured to determine network coverage conditions in a specific area by counting the number of user equipments connected to the access hotspot in a period of time;

[0041] A coverage hotspot identification unit communicatively connected to the hole identification unit and configured to calculate a proportion of the number of connected user equipments to the total number of connected user equipments, determine a distribution of the coverage hotspot, and transmit the identification result to the analysis and evaluation module.

[0042] Preferably, the control module comprises:

[0043] An evaluation result receiving unit is configured to receive the coverage evaluation result sent by the analysis evaluation module, and parse the coverage hole factor, the overlapping coverage factor, the hot area coverage factor and the user perception factor in the coverage evaluation result.

[0044] A control strategy generating unit is in communication connection with the evaluation result receiving unit, and is configured to determine an antenna parameter adjustment strategy based on the factor values in the coverage evaluation result, and generate an optimization instruction.

[0045] A priority management unit is in communication connection with the control strategy generating unit, and is configured to assign an execution priority to the control instruction according to the severity of the coverage problem and the user influence range.

[0046] An instruction distribution unit is in communication connection with the priority management unit, and is configured to distribute the control instruction to the adaptive module according to the instruction priority.

[0047] Preferably, the adaptive module comprises:

[0048] A position orientation control unit is configured to adjust the antenna direction according to the user distribution, and increase the signal strength from the access point to the user equipment.

[0049] A user control unit is in communication connection with the position orientation control unit, and is configured to adjust the access quantity of the user equipment according to the user quantity, and reduce the system interference.

[0050] An adaptive adjustment unit is in communication connection with the user control unit, and is configured to derive the distribution of high-throughput user equipment according to the user equipment rate statistics, adjust the transmission power, and increase the coverage range of the user equipment.

[0051] An intelligent control unit is in communication connection with the adaptive adjustment unit, and is configured to dynamically adjust the antenna transmission power according to the mean value of the user equipment usage data quantity and the throughput, reduce the power for the user whose usage data quantity is lower than the network average value to improve the power consumption efficiency, and increase the power for the user whose usage data quantity is higher than the average value to ensure the transmission quality.

[0052] Preferably, the system further comprises a deployment architecture, wherein:

[0053] The edge side is deployed in a base station controller or a regional centralized node, and comprises an industrial-grade edge server, which is configured to implement the functions of the user management module and the coverage perception module, and perform the manifold construction and the basic feature extraction.

[0054] The center side is deployed in a regional management center or a cloud data center, and comprises a high-performance server cluster, which is configured to implement the functions of the manifold convolution network unit and the coverage evaluation decision unit in the analysis evaluation module.

[0055] The edge side and the center side are connected through a high-reliability low-latency network to realize data exchange and instruction transmission;

[0056] A layered computing strategy is adopted between the edge side and the center side, wherein critical real-time computing is completed on the edge side, and complex analysis computing is executed on the center side.

[0057] The beneficial effects of the present application include:

[0058] 1. By introducing the theory of differential geometry to establish a Riemannian manifold model, the signal propagation characteristics in a complex environment can be accurately expressed, the coverage evaluation accuracy is improved by more than 35%, and the accuracy of cavity identification is improved from 78% of the traditional method to more than 95%;

[0059] 2. Based on the deep learning evaluation model of the manifold convolution network, geometric perception learning is realized, the network model can directly utilize the geometric characteristics of signal propagation for feature extraction, avoiding the loss of geometric information in the traditional method, and the model convergence speed is improved by 3 times;

[0060] 3. The multi-factor coverage evaluation system comprehensively quantifies the network coverage quality, comprehensively considers four dimensions of cavity coverage, overlapping coverage, hot spot coverage and user perception, so that the evaluation result is more comprehensive and stereoscopic;

[0061] 4. The adaptive antenna parameter adjustment mechanism can dynamically optimize network coverage according to user distribution and use behavior, and actual network deployment tests show that the system can improve network coverage rate by 8% to 15% and reduce coverage cavity area by more than 30%;

[0062] 5. The full-automatic coverage optimization closed loop reduces the demand for manual intervention by 80%, the response time to emergencies is shortened from hours to minutes, and the intelligent power control strategy reduces the energy consumption of the base station by 25%. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 is the overall architecture schematic diagram of the intelligent perception wireless network dynamic coverage optimization system of the present application;

[0064] Figure 2 is the structure schematic diagram of the analysis and evaluation module of the present application;

[0065] Figure 3 is the working flow schematic diagram of the manifold construction unit of the present application;

[0066] Figure 4 is the structure schematic diagram of the manifold convolution network unit of the present application;

[0067] Figure 5 is the working flow schematic diagram of the multi-factor coverage evaluation decision unit of the present application;

[0068] Figure 6 is a functional structure diagram of the adaptive module of the present application;

[0069] Figure 7 is a system deployment architecture diagram of the present application. DETAILED DESCRIPTION

[0070] Please refer to Figures 1-7 , the present application will be further described in detail below in combination with the drawings and examples.

[0071] Referring to Figure 1 , the intelligent perception wireless network dynamic coverage optimization system provided by the present application includes a user management module 10, a coverage perception module 20, an analysis and evaluation module 30, a control module 40 and an adaptive module 50. These five modules are connected through communication to form a complete closed-loop control system, realizing intelligent perception, accurate evaluation and dynamic optimization of wireless network coverage.

[0072] The user management module 10 is in communication connection with the coverage perception module 20, used for receiving user equipment access requests, collecting user equipment location information, identifying user equipment types, and transmitting the collected user data to the analysis and evaluation module 30. The coverage perception module 20 is in communication connection with the analysis and evaluation module 30, used for detecting user equipment data fluctuations, identifying access hotspots and uncovered areas, and transmitting the detection results to the analysis and evaluation module 30. The analysis and evaluation module 30 is in communication connection with the control module 40, serving as the core processing unit of the system, responsible for analyzing and evaluating network coverage based on differential geometry theory and deep learning technology, and generating coverage evaluation results. The control module 40 is in communication connection with the adaptive module 50, used for formulating control strategies and generating control instructions according to the evaluation results. The adaptive module 50 receives the control instructions and performs antenna parameter adjustment, realizing dynamic optimization of network coverage.

[0073] Referring to Figure 2 , the analysis and evaluation module 30 is the core processing unit of the system, realizing accurate evaluation of wireless network coverage based on differential geometry theory and deep learning technology. This module includes a manifold construction unit 31, a geometric feature extraction unit 32, a manifold convolution network unit 33 and a coverage evaluation decision unit 34.

[0074] As shown in Figure 3 , the manifold construction unit 31 is used for mapping user equipment location information into a Riemannian manifold, providing a geometric basis for subsequent analysis. This unit includes a data preprocessing module 311, an adjacency relationship construction module 312, a metric tensor calculation module 313 and a manifold generation module 314.

[0075] The data preprocessing module 311 first filters the received user equipment location information for noise and removes outliers. In actual applications, user location data often has errors and outliers, and these abnormal data will affect the accuracy of subsequent modeling. Preferably, the 3σ criterion is used for outlier detection, that is, data points deviating from the mean by more than 3 standard deviations are considered outliers. At the same time, the geographic coordinates are converted into standardized coordinates to generate a set of standardized location points . The standardization process usually uses the Min-Max method to map the data in each dimension to the [0, 1] interval, ensuring the comparability of different dimension data.

[0076] The adjacency relationship construction module 312 calculates the spatial relationship between points based on the set of standardized location points, determines the neighborhood range, and constructs the adjacency matrix A. In an embodiment of the present application, the adaptive K- nearest neighbor algorithm is used to determine the neighborhood range, that is, for each point p in the point set, the value of K is adaptively selected according to the local point density, a smaller K value (such as K = 5) is selected in high-density areas, and a larger K value (such as K = 15) is selected in low-density areas, to balance connectivity and sparsity. In the adjacency matrix A, when point i and point j are adjacent points, A[i,j] = 1; otherwise, A[i,j] = 0.

[0077] The metric tensor calculation module 313 calculates the local metric tensor G for each point in the set of standardized location points. The metric tensor is a key property of Riemannian manifold, reflecting the propagation characteristics of signals in different directions. In the present application, the metric tensor G(p) is a 3x3 symmetric positive definite matrix, and its calculation formula is:

[0078]

[0079] where: G(p) is the metric tensor matrix at point p, which is a 3x3 symmetric positive definite matrix; gij(p) represents the metric component at point p, which represents the metric relationship between the i-th dimension and the j-th dimension, and is determined by the signal strength gradient and user density; the subscripts i and j take the values of 1, 2, and 3, corresponding to the three coordinate axis directions in three-dimensional space. Due to the symmetry property, there is .

[0080] Specifically, the calculation formula of the metric component is:

[0081]

[0082] where: gi(p) represents the signal strength gradient in the i-th direction at point p, with a unit of dB / m, representing the rate of change of signal strength in that direction; gj(p) represents the signal strength gradient in the j-th direction at point p, with a unit of dB / m; ​​is the Kronecker delta function, which is 1 when i = j and 0 otherwise, is a dimensionless judgment function; represents the user density around point p, with the unit of user number / m², usually calculated within a radius of 50 m; and are weight coefficients, both of which are dimensionless quantities, used to balance the influence of signal gradient and user density, and the empirical values are 0.7 and 0.3, respectively. This design of the metric tensor can effectively capture the anisotropic characteristics of signal propagation and has good expression ability for signal attenuation caused by obstacles such as buildings.

[0083] The manifold generation module 314 combines the standardized position point set, the adjacency matrix, and the metric tensor into a discrete Riemannian manifold M = (P, A, G). In addition, a multi-scale manifold representation is constructed to support different accuracy analysis requirements. In practice, usually 3-4 scales of manifold representation are constructed, and the finest scale retains the original accuracy, and the coarse scale reduces the number of points by 30%-50% through downsampling to balance the calculation efficiency and representation accuracy.

[0084] The geometric feature extraction unit 32 is used to extract geometric and topological features from the discrete Riemannian manifold to provide input for the subsequent deep learning model. This unit performs the following key processing:

[0085] First, the Gaussian curvature K(p) and the mean curvature H(p) of each point on the manifold are calculated. The Gaussian curvature is an intrinsic property of the manifold, reflecting the essential bending degree of the covered area; the mean curvature reflects the external bending degree of the covered area. On a discrete manifold, the Gaussian curvature and the mean curvature can be approximated by the following methods:

[0086] ,

[0087] Wherein: K(p) is the Gaussian curvature at point p, with the unit of m⁻²; is the internal angle of the adjacent triangle at point p, with the unit of radian; represents the sum of the internal angles of all n adjacent triangles around point p; is the hybrid Voronoi region area of point p, with the unit of m²; is the number of triangles around point p. The calculation formula of the mean curvature is: ,

[0088]

[0089] ,

[0090] Wherein: H(p) is the mean curvature at point p, with the unit of m​​​-1 ; n is the number of neighbors of point p; denotes the sum of all n neighbors of point p; is the edge length connecting p and neighbor in meters; denotes the angle between the normal vector at point p and the normal vector at point in radians; is the Euclidean distance between point p and point in meters; the fraction is a normalization factor.

[0091] Secondly, geodesic features are extracted and a geodesic distance matrix D between key points is calculated. Geodesic is the shortest path between two points on a manifold, corresponding to the optimal propagation path of a signal. On a discrete manifold, geodesic distance is calculated by Dijkstra algorithm or Fast Marching Method. Experience shows that on a manifold with 100,000 nodes, the optimized Fast Marching Method can complete single-source shortest path calculation within 200 milliseconds, meeting the real-time processing requirements.

[0092] In addition, topological feature extraction is performed, a simple complex on the manifold is constructed, Betti numbers of different dimensions are calculated, and topological defects of the coverage area are identified. Betti numbers and represent the number of connected components, the number of loops, and the number of voids, respectively. In network coverage analysis, indicates that there are multiple non-connected coverage areas, indicates that there is a ring-shaped coverage pattern, then directly indicates the existence of coverage voids. In practice, a well-covered network should generally maintain (single connected) and (no voids).

[0093] Finally, geometric features and topological features are fused into a unified feature vector, and after standardization, the final feature set is generated. Preferably, principal component analysis (PCA) is used for feature dimension reduction, and principal components that explain more than 90% of the variance are retained, usually reducing the feature dimension from the original hundreds of dimensions to 64-128 dimensions.

[0094] Referring to Figure 4The manifold convolution network unit 33 is one of the key innovations of the present application, and is used for deep feature learning on a Riemannian manifold. Traditional convolutional neural networks are mainly designed for regular grid data in Euclidean space and cannot effectively process non-Euclidean geometric data on a manifold. The present application designs a convolutional network structure suitable for a Riemannian manifold, including an input layer 331, a manifold convolution layer 332, a manifold pooling layer 333, a manifold attention layer 334, and a multi-scale feature fusion layer 335.

[0095] The input layer 331 receives the geometric feature vector and the topological feature vector, and organizes the features into a data structure that supports manifold operations. In implementation, a graph data structure is used to store the manifold features, and each node contains local geometric features (such as curvature, geodesic distance) and topological features (such as homology features).

[0096] The manifold convolution layer 332 defines a convolution kernel based on the geodesic distance, and adaptively adjusts the convolution kernel parameters according to the local geometric characteristics. In the present application, the manifold convolution operation is defined as:

[0097] ,

[0098] wherein: is the output feature vector at point p; is the input feature vector at point q; is the geodesic neighborhood of point p, representing the set of all points within a preset threshold of the geodesic distance from point p; represents the summation of all points q in the geodesic neighborhood of point p; is the geodesic distance between point p and point q, with a unit of m; is a weight function based on the geodesic distance; is an adaptive modulation factor considering local geometric characteristics, which is a dimensionless modulation coefficient.

[0099] The weight function is usually a Gaussian kernel, and its formula is:

[0100] ,

[0101] wherein: is the weight value when the distance is d, which is dimensionless; e is the base of the natural logarithm, approximately equal to 2.71828; d is the geodesic distance, with a unit of m; is a learnable parameter that controls the effective range of the convolution kernel, with a unit of m; represents the square of the distance; represents the exponential function, which rapidly decays with the increase of the distance.

[0102] The modulation factor is related to the local curvature of the manifold, and its calculation formula is:

[0103] ,

[0104] where: is the modulation factor between point p and point q, dimensionless; denotes the exponential function with base e; is the Gaussian curvature of point p, with unit of m⁻²; ; is the Gaussian curvature of point q, with unit of m⁻²; ; denotes the absolute difference of Gaussian curvatures of two points; is a learnable parameter, dimensionless, controlling the strength of geometric modulation. Empirical value is = 0.1, which can keep the sensitivity to geometry while not over-amplifying the curvature difference.

[0105] The manifold pooling layer 333 implements a curvature-based adaptive pooling strategy, which keeps high resolution for high-curvature regions and down-samples for low-curvature regions. Specifically, first, the points are sorted by curvature values, and the top 30% of points with the largest curvature values and the remaining 20% of points uniformly sampled are kept, which keeps 50% of points in total, achieving effective dimension reduction while preserving key geometric information.

[0106] The manifold attention layer 334 calculates attention weights based on geometric features, automatically focusing on key areas of network coverage problems. The calculation formula of the attention mechanism is:

[0107] ,

[0108] where: is the normalized attention weight of point p, dimensionless, with a value range of [0, 1]; denotes the exponential function with base e; is the attention score of point p, dimensionless; denotes the summation over all points q on the manifold M; the denominator part is the normalization factor, ensuring that the sum of all attention weights is 1.

[0109] The attention score is calculated by a multi-layer perceptron (MLP) based on local geometric features:

[0110] ,

[0111] where: is the attention score of point p; MLP denotes a multi-layer perceptron, which is a type of feedforward neural network; is the feature vector of point p; is the Gaussian curvature of point p, with unit of m⁻²; is the Gaussian curvature of point q, with unit of m⁻²; the average curvature, in m⁻¹. In practice, a two-layer MLP is usually used with a hidden layer dimension of 64 and a ReLU activation function.

[0112] The multi-scale feature fusion layer 335 constructs a manifold feature pyramid to fuse local and global coverage information at different scales. In an embodiment of the present application, a 3-level feature pyramid is constructed, corresponding to the original scale (100%), the medium scale (50%), and the coarse scale (25%) respectively. The feature fusion adopts a weighted summation method:

[0113]

[0114] wherein: is the fused feature vector; denotes summation over the 3 scale levels; is the weight of the i-th level feature, dimensionless, satisfying ; is the feature vector of the i-th level pyramid. It is initially set to equal weight and is optimized through backpropagation. Practice shows that the optimized weight is usually , , , reflecting the importance of detailed local information to coverage evaluation.

[0115] As shown in Figure 5 , the coverage evaluation decision unit 34 calculates four key coverage factors based on the multi-scale features extracted by the manifold convolution network, realizing comprehensive evaluation of network coverage quality.

[0116] The hole factor F1 represents the network coverage hole situation, and the calculation formula is:

[0117]

[0118] wherein: is the hole factor, dimensionless, with a value range of [0, 1]; denotes the number of user equipment successfully connected to the network, in units of sets; denotes the total number of user equipment, in units of sets. The larger the value, the higher the network coverage hole rate and the worse the coverage quality. According to industry experience, when <0.5, it is determined that the coverage requirement is not met.

[0119] The overlap factor F2 represents the network overlapping coverage situation, and the calculation formula is:

[0120]

[0121] ​​​Wherein: F2 is an overlap factor, dimensionless, the value range is [0, +∞];A represents the number of user equipment of overlapping coverage, unit: set; The number of user equipment representing successful connection network, unit: set. The greater the F2 value, the more serious the terminal overlapping coverage, the network load is unbalanced. Practice shows that when F2≥0.5, it is determined that the requirement is not met, and the antenna parameter needs to be optimized to reduce overlapping coverage.

[0122] The hot area factor F3 measures the hot spot coverage efficiency, and the calculation formula is:

[0123] ,

[0124] Wherein: F3 is a hot spot factor, dimensionless, the value range is [0, 1]; The number of user equipment representing being covered by hot spot, unit: set;A represents the total number of user equipment covered by hot spot, unit: set. The smaller the F3 value, the lower the hot spot coverage efficiency. In actual application, when F3<0.5, it is determined that the requirement is not met, and the hot spot coverage strategy needs to be adjusted.

[0125] The user perception factor F4 evaluates the actual experience of users, and the calculation formula is:

[0126] ,

[0127] Wherein: F4 is a user perception factor, dimensionless, the value range is [0, 1];Rs represents the average throughput of user equipment, unit: Mbps;T represents the measurement time, unit: second;V represents the average rate of regional user equipment, unit: Mbps;T×V represents the data transmission amount that can be reached in theory in the measurement time. The F4 value reflects the ratio of the actual performance of the network to the theoretical performance, and when F4<0.5, it is determined that the requirement is not met, and the user experience is poor.

[0128] In the embodiment of the application, the four factors are comprehensively evaluated in a weighted manner:

[0129] ,

[0130] Wherein: Score is a comprehensive score, dimensionless, the value range is [0, 1]; The weight of each factor, dimensionless, satisfies And Indicates that the poor index is converted into a good index, so that all component directions are consistent, and the greater the value is, the better. According to different scene requirements, the weight is dynamically adjusted. In general urban environment, the experience value is In high-density hot spot area, it can be adjusted to , , more attention is paid to hot spot coverage and overlapping optimization.

[0131] The final evaluation result contains four factor values, a comprehensive score, problem area coordinates, optimization suggestions, and other information, which are sent to the control module 40 for decision-making.

[0132] The user management module 10 is the data entry of the system, responsible for user access authentication, location data processing, and user classification analysis. This module includes a user access authentication unit, a user location processing unit, and a user classification analysis unit.

[0133] The user access authentication unit uses the username / password to authenticate user devices through the Basic authentication method in the HTTP protocol. This authentication method is simple and efficient, suitable for most wireless network environments. After authentication, basic information such as user device identification and access time is recorded.

[0134] The user location processing unit obtains user device geographic location information through various positioning technologies. Preferably, a hybrid positioning strategy is adopted: mainly relying on GPS positioning in outdoor environments, with an accuracy of 5-10 meters; in indoor or weak GPS signal areas, supplemented by LBS information provided by the cellular mobile network or combined with WiFi fingerprint positioning, the integrated positioning error is controlled within 500 meters. Location information is usually represented by latitude and longitude coordinates, with altitude (if available) recorded to form three-dimensional location data.

[0135] The user classification analysis unit divides user devices into three categories: ordinary terminal class, multimedia class, and VR class, using different determination standards. For ordinary terminal class, determination is mainly based on data usage, with a typical threshold of less than 50MB per hour; for multimedia class and VR class, traffic size and throughput are combined for differentiation, where multimedia class devices typically have data usage between 50-500MB per hour and require 2-10Mbps throughput; VR class devices typically have data usage exceeding 500MB per hour and require throughput greater than 10Mbps and are sensitive to latency. This classification helps to implement differentiated coverage optimization strategies for different types of users.

[0136] The coverage awareness module 20 is responsible for detecting network coverage conditions and identifying problem areas. This module includes a device detection unit, a hole identification unit, and a coverage hotspot identification unit.

[0137] The device detection unit detects the connection and online status of user devices to the access hotspot by periodically sending ICMP echo request (ping) and receiving ICMP echo response. In practice, ping detection is performed every 5 minutes for each access user device, the timeout time is set to 300 ms, and 3 consecutive non-responses are determined as connection abnormality. According to the online status of the user device, it is judged whether the network coverage is blocked. This active detection method more accurately reflects the actual connection state than passive signal monitoring.

[0138] The coverage hole identification unit determines the network coverage condition in a specific area by counting the number of user devices accessing the hotspot in a period of time. In specific implementation, the coverage area is divided into grids (usually 100m x 100m), and the trend of the number of access devices in each grid is calculated. If the number of access devices in a certain grid is less than 30% of the surrounding grids and lasts for more than 30 minutes, it is preliminarily determined that there is a coverage hole in that area. Preferably, time series analysis is combined with historical data to exclude false hole identification caused by temporary changes in human flow.

[0139] The coverage hotspot identification unit calculates the proportion of access users to the total number of connected users to determine the distribution of coverage hotspots. Hotspot identification uses a density clustering algorithm, which first calculates a user density heat map and then identifies areas with a density more than twice the average as potential hotspots. For the identified hotspot area, further analyze the proportion of access users: if the proportion of access users in a certain hotspot area to the total number of connected users exceeds 20%, mark it as a key coverage hotspot and prioritize its coverage quality.

[0140] The control module 40 is a bridge connecting the analysis and evaluation module 30 and the adaptive module 50, responsible for generating control instructions according to the evaluation results. This module includes an evaluation result receiving unit, a control strategy generating unit, a priority management unit and an instruction distribution unit.

[0141] The evaluation result receiving unit receives the coverage evaluation results sent by the analysis and evaluation module 30, and parses the four key factor values (hole factor F1, overlapping coverage factor F2, hot area factor F3 and user perception factor F4) and problem area coordinates in the information.

[0142] The control strategy generating unit determines the antenna parameter adjustment strategy based on the factor values in the evaluation results. The strategy generation uses a combination of rule engine and heuristic algorithm, generating corresponding optimization instructions for different factor value combinations. For example, when F1<0.5 (hole problem), generate an instruction to increase the coverage range of the antenna; when F2≥0.5 (overlapping problem), generate an antenna parameter adjustment instruction to reduce overlapping coverage; when F3<0.5 (hotspot coverage insufficient), generate a hotspot coverage enhancement instruction; when F4<0.5 (poor user experience), generate an instruction to improve signal quality.

[0143] The priority management unit assigns execution priority to the control instructions according to the severity of the coverage problem and the user impact range. The priority is usually divided into four levels: emergency (to be executed within 5 minutes), high (to be executed within 30 minutes), medium (to be executed within 2 hours), and low (to be executed within 24 hours). In a typical priority assignment strategy, the case where the number of users affected exceeds 1000 and F4 < 0.3 is marked as emergency; the case where the number of users affected is between 500 and 1000 or 0.3 ≤ F4 < 0.4 is marked as high priority; and other cases are assigned medium or low priority according to specific parameters.

[0144] The instruction distribution unit distributes the control instructions to the adaptive module 50 according to the instruction priority. The instruction distribution adopts an asynchronous message queue mechanism to ensure that high-priority instructions are executed first while avoiding instruction conflicts. For instructions that may conflict (such as adjusting different parameters of the same antenna at the same time), the system performs conflict detection and merging processing to generate coordinated composite instructions.

[0145] Referring to Figure 6 The adaptive module 50 is responsible for executing the control instructions sent by the control module 40 to make actual adjustments to the network coverage. This module includes a position orientation control unit 51, a user control unit 52, an adaptive adjustment unit 53, and an intelligent control unit 54.

[0146] The position orientation control unit 51 adjusts the antenna direction according to the user distribution to increase the signal strength from the access point to the user equipment. In implementation, this unit first generates a user density heat map based on user location data to identify user-intensive areas; then calculates the angular deviation between the current antenna direction and the user-intensive areas; and finally generates an antenna azimuth adjustment instruction to align the antenna main lobe direction with the user-intensive areas. The adjustment accuracy is usually 5 degrees, and the maximum adjustment range is ±30 degrees to avoid affecting the coverage of adjacent cells.

[0147] The user control unit 52 adjusts the number of user equipment access according to the number of users to reduce system interference. When it is detected that the number of users of an access point is close to saturation (usually 85% of the theoretical capacity), a load sharing mechanism is started: appropriately reducing the signal strength of the access point, while increasing the signal strength of the surrounding low-load access points to guide new users to the low-load access points; for existing users, it is determined whether to perform forced switching according to the service priority and signal quality evaluation. Practice shows that this mechanism can reduce the load of high-load access points by 15%-25% while maintaining user experience quality.

[0148] The adaptive adjustment unit 53 obtains the distribution of high-throughput user equipment according to user equipment rate statistics, adjusts the transmission power, and increases the coverage of user equipment. The unit performs user equipment rate statistics every 15 minutes, identifies the high-throughput user concentration area, and increases the coverage power accordingly. The increase is usually 1-3 dB, and the maximum is not more than 5 dB, to avoid excessive interference to adjacent cells.

[0149] The intelligent control unit 54 dynamically adjusts the antenna transmission power according to the mean value of the user equipment usage data volume and the throughput, to realize fine power control. Specifically, for the area where the user equipment usage data volume is lower than the network average, the transmission power is appropriately reduced (usually by 1-2 dB), to improve energy utilization efficiency; for the area where the user equipment usage data volume is higher than the average, the transmission power is appropriately increased (usually by 1-3 dB), to ensure transmission quality. This differentiated power control strategy meets the needs of different users and optimizes energy utilization efficiency, and the actual measurement can reduce the energy consumption of the base station by 20%-30%.

[0150] As shown in Figure 7 The system of the present application adopts an edge-center collaborative deployment architecture, fully utilizes distributed computing resources, and improves system efficiency and reliability.

[0151] The edge side is deployed in a base station controller or a regional centralized node, equipped with an industrial-grade edge server, usually with an 8-core CPU, 16 GB of memory and GPU acceleration capability. The edge side mainly realizes the functions of the user management module 10 and the coverage awareness module 20, and simultaneously performs the manifold construction and basic feature extraction in the analysis and evaluation module 30, which usually requires low-latency processing and is suitable for execution on the edge side close to the data source.

[0152] The center side is deployed in a regional management center or a cloud data center, equipped with a high-performance server cluster, supporting large-scale parallel computing. The center side mainly realizes the functions of the manifold convolution network unit 33 and the coverage evaluation decision unit 34 in the analysis and evaluation module 30, and the functions of the control module 40, which have high computational complexity and are suitable for execution on the center side with abundant resources.

[0153] The edge side and the center side are connected through a high-reliability low-latency network, usually using a dedicated network or a 5G slice to ensure communication quality, with an end-to-end delay controlled within 20 ms. A layered computing strategy is adopted between the two sides, with key real-time computing completed on the edge side and complex analysis computing performed on the center side, to realize efficient utilization of computing resources.

[0154] In addition, the system supports multi-region deployment and disaster recovery switching, to ensure the reliability of the service. When a regional computing node fails, the adjacent regional node can take over its tasks to ensure the continuous operation of the system.

[0155] The following is a typical workflow example to illustrate the cooperative working mode of each module of the system of the present application:

[0156] 1. Initial data collection phase: User management module 10 receives user device access request and performs authentication, collects user device location information and usage data, and at the same time, coverage awareness module 20 collects network coverage information through active detection and passive listening. The two modules transmit the collected data to analysis and evaluation module 30.

[0157] 2. Manifold modeling phase: The manifold construction unit 31 of the analysis and evaluation module 30 receives user location data, and after preprocessing, it constructs a Riemannian manifold model. The geometric feature extraction unit 32 calculates the curvature feature, geodesic feature and topological feature on the manifold, and generates a feature vector set.

[0158] 3. Deep learning evaluation phase: The manifold convolution network unit 33 receives the feature vector set, and through the manifold convolution layer, pooling layer and attention layer, it performs feature extraction and fusion, and outputs high-level semantic features. The coverage evaluation decision unit 34 calculates four coverage factors based on these features, and generates coverage evaluation results.

[0159] 4. Control decision phase: Control module 40 receives coverage evaluation results, formulates control strategies according to four factor values, allocates execution priority, generates control instructions and sends them to adaptive module 50.

[0160] 5. Parameter adjustment phase: Adaptive module 50 adjusts antenna position, direction, transmission power and gain parameters according to control instructions to optimize network coverage. At the same time, it feeds back the adjustment results to analysis and evaluation module 30 to form a closed-loop control.

[0161] In a practical application case, the system is deployed in a large commercial area of a certain city, covering an area of about 5 square kilometers, including 30 base stations and 150 cells. Within one month after deployment, the regional network coverage rate increased from 85% to 96%, the coverage hole area decreased by 42%, the user complaint rate decreased by 58%, and the network energy consumption decreased by 23%. Especially in crowded areas such as shopping malls and subway stations, the system can adjust the coverage strategy in real time according to the flow changes, and the network experience satisfaction rate in peak hours increased from 73% to 92%.

[0162] In summary, the intelligent sensing wireless network dynamic coverage optimization system provided by the present application introduces differential geometry theory and deep learning technology, realizes intelligent sensing, accurate evaluation and dynamic optimization of wireless network coverage, greatly improves network coverage quality and user experience, and at the same time reduces energy consumption and operation and maintenance cost, has significant technical advantages and application value.

[0163] The above merely provides the preferred embodiments of the present application, and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall fall into the scope of protection of the present application.

Claims

1. A system for intelligent sensing based dynamic coverage optimization for wireless networks, characterized in that, The application relates to a network coverage evaluation system, comprising: a user management module for receiving user equipment access requests and performing authentication, collecting user equipment location information, identifying the types of user equipment, and transmitting the collected user data to an analysis and evaluation module; a coverage perception module in communication connection with the user management module, for detecting user equipment data fluctuations, identifying access hotspots and uncovered areas, and transmitting the detection results to the analysis and evaluation module; an analysis and evaluation module in communication connection with the user management module and the coverage perception module, for mapping the user equipment location information into a Riemann manifold, extracting geometric features and topological features of the Riemann manifold, constructing a manifold convolutional neural network through a manifold convolution layer for extracting spatial features, a manifold pooling layer for dimension reduction processing, and a manifold attention mechanism for identifying key areas, realizing adaptive pooling based on curvature, maintaining high resolution for high-curvature areas and reducing sampling for low-curvature areas, calculating coverage cavity factors, overlapping coverage factors, hot area coverage factors and user perception factors according to the geometric features and topological features, generating coverage evaluation results, and sending the coverage evaluation results to a control module; a control module in communication connection with the analysis and evaluation module, for receiving the coverage evaluation results, determining a control strategy based on the coverage evaluation results, generating control instructions, and sending the control instructions to an adaptive module; an adaptive module in communication connection with the control module, for receiving the control instructions, adjusting the positions and directions of antennas according to the control instructions, dynamically adjusting the transmitting power and antenna gain, and feeding back the adjustment results to the analysis and evaluation module.

2. The intelligent, sensory, wireless network dynamic coverage optimization system of claim 1, wherein, The analysis and evaluation module comprises: a manifold construction unit for receiving the user equipment location information, pre-processing user location data, constructing an adjacency relationship matrix, calculating a local metric tensor, and generating a discrete Riemann manifold; a geometric feature extraction unit in communication connection with the manifold construction unit, for calculating the Gaussian curvature and average curvature on the discrete Riemann manifold, extracting geodesic line features, analyzing the topological structure of the discrete Riemann manifold, and generating a geometric feature vector and a topological feature vector; a manifold convolution network unit in communication connection with the geometric feature extraction unit, for receiving the geometric feature vector and the topological feature vector, extracting spatial features through a manifold convolution layer, dimension reduction processing through a manifold pooling layer, applying a manifold attention mechanism to identify key areas, and fusing multi-scale feature information; a coverage evaluation decision unit in communication connection with the manifold convolution network unit, for calculating coverage cavity factors, overlapping coverage factors, hot area coverage factors and user perception factors based on the multi-scale feature information, comprehensively evaluating network coverage quality, and generating the coverage evaluation results.

3. The intelligent, sensory, wireless network dynamic coverage optimization system of claim 2, wherein, The manifold construction unit comprises: a data pre-processing module for filtering noise and removing outliers from the user equipment location information, converting geographic coordinates into standardized coordinates, and generating a standardized location point set; an adjacency relationship construction module in communication connection with the data pre-processing module, for calculating the spatial relationship between points based on the standardized location point set, determining a neighborhood range, and constructing an adjacency matrix. A metric tensor calculation module, in communication connection with the adjacency relation construction module, is configured to calculate a local metric tensor for each point in the standardized position point set, the local metric tensor reflecting anisotropic characteristics of signal propagation; A manifold generation module, in communication connection with the metric tensor calculation module, is configured to combine the standardized position point set, the adjacency matrix and the local metric tensor into a discrete Riemannian manifold, and construct a multi-scale manifold representation.

4. The intelligent, sensory, wireless network dynamic coverage optimization system of claim 2, wherein, The manifold convolution network unit comprises: An input layer configured to receive the geometric feature vector and the topological feature vector, and organize the features into a data structure supporting manifold operation; A manifold convolution layer, in communication connection with the input layer, is configured to define a convolution kernel based on geodesic distance, and adaptively adjust the convolution kernel parameters according to local geometric characteristics, and extract local features; A manifold pooling layer, in communication connection with the manifold convolution layer, is configured to realize adaptive pooling based on curvature, and maintain high resolution for high-curvature areas and down-sample low-curvature areas; A manifold attention layer, in communication connection with the manifold pooling layer, is configured to calculate attention weights based on geometric features, and automatically focus on key areas of network coverage problems; A multi-scale feature fusion layer, in communication connection with the manifold attention layer, is configured to construct a manifold feature pyramid, fuse local and global coverage information of different scales, and generate a comprehensive feature representation.

5. The intelligent, sensory, wireless network dynamic coverage optimization system of claim 2, wherein, The coverage evaluation decision unit calculates the coverage factor in the following manner: Cavity factor wherein denotes the number of devices of the user equipment successfully connected to the network, denotes the total number of devices of the user equipment; Overlap factor where A represents the number of user equipments covered by the overlap. Hot zone factor wherein represents the number of user equipments covered by the hotspot, A represents the total number of user equipments covered by the hotspot; User perception factor wherein T denotes the average throughput of the user equipment, T denotes the measurement time, and V denotes the average velocity of the user equipment in the area. When F1<0.5, F2≥0.5, F3<0.5 or F4<0.5, it is determined that the network coverage requirement is not met.

6. The intelligent, sensory, wireless network dynamic coverage optimization system of claim 1, wherein, The user management module comprises: A user access authentication unit configured to authenticate user equipment through a Basic authentication method in an HTTP protocol using a username / password; A user location processing unit, in communication connection with the user access authentication unit, is configured to obtain geographic location information of the user equipment through GPS positioning technology, calibrate the location information through LBS information provided by a cellular mobile network or a third-party map platform, and control the positioning error to be no more than 500 meters; A user classification analysis unit, in communication connection with the user location processing unit, is configured to divide the user equipment into ordinary terminal class, multimedia class and VR class, wherein the ordinary terminal class user equipment uses data volume as a judgment basis, and the multimedia class and VR class user equipment use traffic size and throughput as a distinguishing basis.

7. The intelligent, perception-based, wireless network dynamic coverage optimization system of claim 1, wherein, The coverage perception module comprises: A device detection unit configured to detect connection and online status of user equipment and access hotspots by periodically sending ICMP echo request and receiving ICMP echo response, and determine whether the network coverage is blocked according to the online status of the user equipment; A hole identification unit, in communication connection with the device detection unit, is configured to determine network coverage conditions in a specific area by counting the number of user equipment accessing the hotspots within a period of time; A coverage hotspot identification unit, in communication connection with the hole identification unit, is configured to calculate the proportion of access users in the total connected users, determine the distribution of coverage hotspots, and transmit the identification result to the analysis and evaluation module.

8. The intelligent, perception-based, wireless network dynamic coverage optimization system of claim 1, wherein, The control module comprises: An evaluation result receiving unit is configured to receive the coverage evaluation result sent by the analysis evaluation module, and parse the coverage hole factor, the overlapping coverage factor, the hot area coverage factor and the user perception factor in the coverage evaluation result; A control strategy generating unit is in communication connection with the evaluation result receiving unit, and is configured to determine an antenna parameter adjustment strategy based on the factor values in the coverage evaluation result, and generate an optimization instruction; A priority management unit is in communication connection with the control strategy generating unit, and is configured to assign an execution priority to the control instruction according to the severity of the coverage problem and the user influence range; An instruction distribution unit is in communication connection with the priority management unit, and is configured to distribute the control instruction to the adaptive module according to the instruction priority.

9. The intelligent, perception-based, wireless network dynamic coverage optimization system of claim 1, wherein, The adaptive module comprises: A position orientation control unit is configured to adjust the antenna direction according to the user distribution, and increase the signal strength from the access point to the user equipment; A user control unit is in communication connection with the position orientation control unit, and is configured to adjust the access quantity of the user equipment according to the user quantity, and reduce the system interference; An adaptive adjustment unit is in communication connection with the user control unit, and is configured to obtain the distribution of the high-throughput user equipment according to the user equipment rate statistics, adjust the transmission power, and increase the coverage range of the user equipment; An intelligent control unit is in communication connection with the adaptive adjustment unit, and is configured to dynamically adjust the antenna transmission power according to the mean value of the user equipment usage data quantity and the throughput, reduce the power for the user whose usage data quantity is lower than the network average value to improve the power consumption efficiency, and increase the power for the user whose usage data quantity is higher than the average value to ensure the transmission quality.

10. The intelligent, perception-based, wireless network dynamic coverage optimization system of claim 1, wherein, The system further comprises a deployment architecture, in which: The edge side is deployed in a base station controller or a regional centralized node, and comprises an industrial-grade edge server, which is configured to implement the functions of the user management module and the coverage perception module, and perform manifold construction and basic feature extraction; The center side is deployed in a regional management center or a cloud data center, and comprises a high-performance server cluster, which is configured to implement the functions of the manifold convolution network unit and the coverage evaluation decision unit in the analysis evaluation module; The edge side and the center side are connected through a high-reliability low-latency network to realize data exchange and instruction transmission; The edge side and the center side adopt a hierarchical computing strategy, in which the key real-time computing is completed on the edge side, and the complex analysis computing is performed on the center side.

Images