Deep Learning-Based Channel State Information Prediction System for Large-Scale Antenna Arrays

Abstract

This invention discloses a deep learning-based channel state information prediction system for large-scale antenna arrays, relating to the mobile internet field. It addresses the problem of poor prediction performance in existing large-scale antenna array channel state information prediction systems. The system includes a data acquisition module: performing array channel region overlap analysis between the target channel location and historical channel locations; matching the historical and target channel locations based on the analysis results to obtain matching location selection data; a location prediction module: creating a weather signal impact model; predicting overall signal anomalies in the target channel location based on the weather signal impact model to obtain overall location signal prediction data; and a channel early warning module: directly issuing early warnings for locations with signal anomalies; analyzing local signal obstruction in locations with acceptable signal strength; and issuing local early warnings based on the analysis results. This invention improves the accuracy and comprehensiveness of the state information prediction results.

Description

Technical Field

[0001] This invention belongs to the field of mobile Internet and involves deep learning technology, specifically a deep learning-based large-scale antenna array channel state information prediction system. Background Technology

[0002] Existing large-scale antenna array channel state information prediction systems have the following drawbacks when performing signal prediction: 1. Existing large-scale antenna array channel state information prediction systems typically select historical sites where the same type of large-scale antenna array has been deployed as target matching sites for channel information prediction. However, in real-world scenarios, due to differences in geographical environment, building layout, and other geographical parameters among different historical sites, even if the same type of large-scale antenna array has been deployed, their actual operating states will vary significantly. This difference can easily lead to a lack of sufficient common features between matching sites, resulting in incomplete matching indicators for the target matching site and ultimately a significant reduction in the accuracy of channel information prediction results. 2. Most existing large-scale antenna array channel state information prediction systems perform overall channel state prediction and early warning for the target channel site. However, in real-world scenarios, the number of service objects covered by different array units and the distribution of service object locations are often inconsistent in the same channel site. If an overall state prediction method is used, it is difficult to effectively detect local channel state anomalies, which can easily lead to a lack of comprehensiveness in the channel state prediction results.

[0003] To address this, we propose a deep learning-based system for predicting channel state information for large-scale antenna arrays. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this invention is to provide a deep learning-based channel state information prediction system for large-scale antenna arrays. This invention aims to improve the accuracy and comprehensiveness of the channel state information prediction system for large-scale antenna arrays. To achieve the above objectives, the present invention adopts the following technical solution: a deep learning-based large-scale antenna array channel state information prediction system, the specific working process of each module is as follows: Data acquisition module: Performs array cell channel region overlap analysis on the target channel site and the historical channel site. Based on the analysis results, performs site similarity matching between the historical channel site and the target channel site to obtain matching site selection data. Site prediction module: Creates a weather signal impact model based on the matched site selection data, performs overall signal anomaly prediction for the target channel site based on the weather signal impact model, and obtains overall site signal prediction data based on the prediction results; Channel early warning module: It directly issues early warnings for sites with abnormal signals based on the overall signal prediction data of the site, and performs local signal obstruction analysis for sites with qualified signals, and issues local early warnings based on the analysis results.

[0005] Furthermore, the matching site selection data is obtained as follows: Obtain the target channel site and multiple historical channel sites, and arbitrarily select one sample channel site from the multiple historical channel sites. Array channel coverage analysis is performed on the sample channel site and the target channel site, and the overlap degree of the channel area corresponding to the sample channel site is obtained based on the analysis results. Obtain the channel area overlap degree corresponding to each historical channel site and the target channel site, set the channel area overlap degree selection value, if the channel area overlap degree is greater than or equal to the channel area overlap degree selection value, then select the corresponding historical channel site as the target matching site, if it is less than, then select the corresponding historical channel site as a non-target matching site, and obtain the matching site selection data.

[0006] Furthermore, an array channel coverage analysis was performed on the sample channel site, as detailed below: The array elements contained in the large-scale antenna array device are acquired, and a sample array element is randomly selected from the acquired array elements. The deployment area of ​​the sample array unit in the target channel field is set as the target array area, and the deployment area of ​​the sample array unit in the sample channel field is set as the sample array area. Perform signal physical range coverage analysis on the edge points corresponding to the target array area, and obtain the effective path geometric components corresponding to each edge point of the target array area based on the analysis results; Obtain the effective path geometric components corresponding to each edge point of the target array region, sort them in descending order to obtain the path component sorting queue, and divide the path component sorting queue into several path component subsequences with the same interval span.

[0007] Furthermore, an array channel coverage analysis was performed on the sample channel site, as detailed below: In the multiple path component subsequences obtained, arbitrarily select an X1 component subsequence. In the target region planar image, mark the edge points in the X1 component subsequence as X1 edge points. Obtain the geometric center point of the area covered by the sample array unit to obtain the geometric center point of the sample array. Obtain the line connecting the X1 edge points and the geometric center point of the sample array to obtain multiple X1 edge lines. Set the spatial area covered by the obtained X1 edge lines in the target region planar image as the target region of the X1 sequence. Obtain the target region corresponding to each component subsequence to obtain multiple target regions; Perform signal physical range coverage analysis on the edge points corresponding to the sample array area, and obtain the effective path geometric components corresponding to each edge point of the sample array area based on the analysis results. Acquire a planar image of the sample region, mark the edge points in the X1 component subsequence as X1 edge points, acquire the geometric center points of the area covered by the sample array unit in the planar image of the sample region, obtain the geometric center points of the sample array, acquire the lines connecting the X1 edge points and the geometric center points of the sample array, obtain multiple X1 edge lines, and set the spatial area covered by the acquired X1 edge lines in the planar image of the sample region as the X1 sequence sample region; Obtain the sequence sample regions corresponding to different sequence sample regions to obtain multiple sequence sample region values.

[0008] Furthermore, an array channel coverage analysis was performed on the sample channel site, as detailed below: Obtain the area ratio of each sequence target region to the target array region to obtain multiple target area ratios; obtain the area ratio of each sequence sample region to the sample array region to obtain multiple sample area ratios. The channel path region deviation corresponding to the sample array unit is obtained by calculating the target area ratio and the sample area ratio. Set a channel path area deviation benchmark interval. If the channel path area deviation is within the channel path area deviation benchmark interval, the sample array area in the sample channel field is divided into a channel consistent area. If it is not within the benchmark interval, the sample array area in the sample channel field is divided into a channel inconsistent area. The number of consistent channel regions is counted as B1, and the number of inconsistent channel regions is counted as B2. The overlap of channel regions is obtained by calculating B1 / (B1+B2).

[0009] Furthermore, the geometric components of the effective path are obtained, as follows: Acquire a planar image of the target region, mark the sample array units in the planar image of the target region, select a feature signal point, mark the edge of the target array region in the planar image of the target region, and select a feature edge point; In the planar image of the target region, the distance between the lines connecting the feature signal points and the feature edge points is obtained to get the projected distance of the model points. Create a 3D model of the target array region, and mark the vertical planes of feature points and edge points in the model. Collect the vertical distance between the vertical planes of feature points and edge points to obtain the longitudinal distance of the point model. In the 3D model of the target array region, create the reference direct path angle to obtain the longitudinal distance Zjl of the point model and the projected distance Hjl of the model point. Then, use the formula... The actual distance between the site locations is calculated, and the product of the actual distance between the site locations and the cosine of the angle between the actual distance between the site locations and the reference direct path is calculated to obtain the channel path geometric components between the feature edge points and the feature signal points. Obtain the channel path geometric components between the feature edge point and each signal point, and compare the values ​​of the obtained multiple channel path geometric components. Set the channel path geometric component with the smallest value as the effective path geometric component corresponding to the feature edge point.

[0010] Furthermore, the overall signal prediction data for the site is acquired, as follows: Obtain matching site selection data, acquire target matching sites based on matching site selection data, obtain the ratio of cumulative precipitation to activity cycle duration for each target matching site during the activity, obtain multiple site precipitation intensities, and label them from smallest to largest as the first site precipitation intensity to the jth site precipitation intensity. The average time delay of the site signal corresponding to the precipitation intensity of the first site to the precipitation intensity of the j-th site is set as the time delay of the first site signal to the time delay of the j-th site signal, respectively. By performing polynomial fitting on the precipitation intensity from the first site to the precipitation intensity from the j-th site and the signal delay from the first site to the signal delay from the j-th site, a weather signal impact model is obtained. The precipitation intensity at the target site is obtained by the ratio of the predicted precipitation at the target site to the duration of the activity. The precipitation intensity of the target site is input into the weather signal impact model to obtain the predicted value of the site signal delay. A signal delay reference interval is set. If the predicted value of the site signal delay is within the signal delay reference interval, the target channel site is determined to be a signal assessment area. If it is not within the reference interval, the target channel site is determined to be a signal abnormality site. If the target channel site is a signal analysis area, then the channel direct bias analysis is performed on the target channel site. Based on the analysis results, the signal analysis area is further screened to obtain the overall signal prediction data of the site.

[0011] Furthermore, signal anomaly warnings are issued for the target channel site, as detailed below: Obtain overall site signal prediction data, and based on the overall site signal prediction data, identify sites with acceptable signal and sites with abnormal signal. If the target channel site is a channel abnormal site, a site channel abnormality warning will be issued directly. If the target channel site is a channel qualified site, a site local signal obstruction analysis will be performed on the target channel site, and a site local warning will be issued based on the analysis results. The different array regions contained in the target channel site are acquired, and a feature array region is arbitrarily selected from the acquired array regions. Signal occlusion analysis is performed on the feature array region, and the degree of direct signal occlusion corresponding to the feature array region is determined based on the analysis results. Obtain the direct signal obstruction degree corresponding to each array area and set the direct signal obstruction degree benchmark range. If the direct signal obstruction degree is within the direct signal obstruction degree benchmark range, it is determined that the signal obstruction of the corresponding array area is normal. If it is not within the benchmark range, it is determined that the signal obstruction of the corresponding array area is abnormal and a signal obstruction warning is issued.

[0012] Furthermore, signal occlusion analysis is performed on the characteristic array region, as follows: A three-dimensional model is created for the feature array region to obtain the feature region three-dimensional model. In the feature region three-dimensional model, the signal receiving and transmitting point layout plane corresponding to the feature array unit is set as the first region feature plane, and the ground area where the array service object is located is set as the second region feature plane. The first region feature plane is divided into several region pixels. A ray perpendicular to the second region feature plane is drawn through each region pixel to obtain multiple region pixel rays. Then, a sample region pixel ray is randomly selected from the multiple region pixel rays. The model region covered by the pixel ray of the sample region is divided into several ray pixels. The vertical distance between each ray pixel and the ground is obtained to obtain the vertical distance of the pixel.

[0013] Furthermore, signal occlusion analysis is performed on the characteristic array region, as follows: Several array service objects are selected in the target matching area. The maximum lifting height of the terminal device of each array service object in the target matching area is obtained. The average of the obtained maximum lifting height of the terminal device is calculated to obtain the first lifting height feature value. The standard deviation of the obtained maximum lifting height of the terminal device is calculated to obtain the second lifting height feature value. The sum of the first lifting height feature value and the second lifting height feature value is set as the upper limit of the vertical distance target interval. The minimum lifting height of each terminal device in the target matching area is obtained. The average of the obtained minimum lifting height of the terminal device is calculated to obtain the third lifting height feature value. The standard deviation of the obtained minimum lifting height of the terminal device is calculated to obtain the fourth lifting height feature value. The difference between the third lifting height feature value and the fourth lifting height feature value is set as the lower limit of the vertical distance target interval, thus obtaining the vertical distance target interval. Pixels whose vertical distance is within the target vertical distance range are acquired to obtain multiple test pixels. The first region feature plane is divided into several planar pixels. Lines are drawn between each planar pixel and the test pixel to obtain multiple signal pseudo-direct connection paths. If there is no direct path among the acquired multiple signal pseudo-direct connection paths, the test pixel is classified as a non-directly oriented pixel. If any direct path exists among the acquired multiple signal pseudo-direct connection paths, the test pixel is classified as a directly oriented pixel. The types of the pixels to be tested contained in each region pixel ray are classified. The number of pixels to be tested separated by the second region feature plane is counted to obtain the first pixel feature value. The number of the direct-ray pixels separated by the second region feature plane is counted to obtain the second pixel feature value. Calculate the difference between the feature value of the first pixel and the feature value of the second pixel, and obtain the ratio of the obtained difference to the feature value of the first pixel to obtain the direct signal occlusion degree corresponding to the feature array region.

[0014] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention uses historical sites where large-scale antenna arrays of the same type are deployed as target matching sites for channel information prediction, while analyzing the similarity of the environmental conditions of the site area covered by each signal element in the large-scale antenna array for further site matching. This improves the comprehensiveness of the target matching site matching index and further ensures the accuracy of the channel information prediction results. 2. This invention performs overall channel information status prediction and early warning for target channel sites, and screens qualified and abnormal channel sites based on the prediction results. It also performs direct channel obstruction analysis by combining the lifting height range of signal terminal equipment in qualified channel sites, and provides early warning of channel information anomalies in the local environment of the site based on the analysis results, thereby improving the comprehensiveness and pertinence of channel status information prediction results. Attached Figure Description

[0015] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0016] Figure 1 This is an overall system block diagram of the present invention; Figure 2 This is a schematic diagram of the included angle of the reference direct connection path in this invention; Figure 3 This is a schematic diagram of the channel transmit / receive angle of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all 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.

[0018] Example 1 Please see Figure 1 The present invention provides a technical solution: a deep learning-based large-scale antenna array channel state information prediction system, including a data acquisition module, a site prediction module, a channel early warning module and a server. The data acquisition module, the site prediction module and the channel early warning module are respectively connected to the server, and the server controls the data acquisition module, the site prediction module and the channel early warning module respectively. The data acquisition module performs array cell channel region overlap analysis on the target channel site and the historical channel site. Based on the analysis results, it performs site similarity matching between the historical channel site and the target channel site to obtain matching site selection data. Specifically as follows: For large-scale event sites that require large-scale antenna array channel state information prediction, the target channel site is obtained. Historical activity sites that have deployed the same large-scale antenna array equipment as the target channel site are acquired to obtain multiple historical channel sites, and a sample channel site is randomly selected from the multiple acquired historical channel sites. It should be noted here that: In this application, the antenna distribution elements in the large-scale antenna array involved herein are deployed in a distributed manner, that is, the antenna elements are dispersed in multiple physical locations and connected by wireless links; In this application, the large-scale antenna array channel state information referred to herein specifically refers to a site where a large-scale antenna array channel state information prediction system is deployed. In this application, the large event venues referred to herein are specifically densely populated event venues, including but not limited to music festival venues, outdoor concert venues, and outdoor football match venues. Array channel coverage analysis is performed on the sample channel site and the target channel site, and the overlap degree of the channel area corresponding to the sample channel site is obtained based on the analysis results. Specifically as follows: The array elements contained in the large-scale antenna array device are acquired, and a sample array element is randomly selected from the acquired array elements. The deployment area of ​​the sample array unit in the target channel field is set as the target array area, and the deployment area of ​​the sample array unit in the sample channel field is set as the sample array area. It should be noted here that: In this application, the deployment area of ​​the sample array region within the target channel site is specifically the working area where the sample array unit enhances the antenna signal; In this application, both the target array region and the sample array region referred to herein are audience regions; Perform signal physical range coverage analysis on the edge points corresponding to the target array area, and obtain the effective path geometric components corresponding to each edge point of the target array area based on the analysis results; Specifically as follows: The target array region is acquired by performing regional planar image acquisition to obtain a target region planar image. The sample array units are marked in the target region planar image, and a feature signal point is arbitrarily selected in the marked region of the sample array unit. The target array region is marked in the target region planar image, and a feature edge point is arbitrarily selected in the marked region edge. It should be noted here that: In this application, the feature signal points involved are preferably selected from points where the array signal transmission and reception are integrated. If there are points where the signal transmission and reception are not integrated, a feature signal transmission point is arbitrarily selected, and the corresponding signal reception point is obtained to obtain the feature signal reception point. The midpoint of the line connecting the feature signal transmission point and the feature signal reception point is set as the feature signal point.

[0019] In the planar image of the target region, the distance between the line connecting the feature signal points and the feature edge points is obtained to obtain the model point projection distance; A spatial three-dimensional model of the target array region is created to obtain the three-dimensional model of the target array region. In the three-dimensional model of the target array region, a plane perpendicular to the ground is drawn through the feature signal points to obtain the vertical plane of the feature points. A plane perpendicular to the ground is drawn through the feature edge points to obtain the vertical plane of the edge points. The vertical distance between the vertical plane of the feature points and the vertical plane of the edge points is collected to obtain the longitudinal distance of the point model. Please see Figure 2In the three-dimensional model of the target array area, the signal transmission and reception point layout plane corresponding to the sample array unit is set as the array layout plane. A ray perpendicular to the array layout plane is drawn through the feature signal point to obtain the feature point signal baseline. The line connecting the feature signal point and the feature edge point is set as the signal direct transmission path marker line. The angle between the feature point signal baseline and the signal direct transmission path marker line at the feature signal point is obtained to obtain the baseline direct connection path angle. It should be noted here that: In this application, the signal transmission and reception points involved include signal transmission points and signal reception points.

[0020] Convert the longitudinal distance of the point model and the projected distance of the model point into the actual distance on the site to obtain the longitudinal distance Zjl of the point model and the projected distance Hjl of the model point. Then, use the formula... The actual distance between the site locations is calculated, and the product of the actual distance between the site locations and the cosine of the angle between the actual distance between the site locations and the reference direct path is calculated to obtain the channel path geometric components between the feature edge points and the feature signal points. Repeat the process of obtaining the channel path geometric components between the feature edge point and the feature signal point, obtain the channel path geometric components between the feature edge point and each signal point respectively, and compare the values ​​of the obtained multiple channel path geometric components. Set the channel path geometric component with the smallest value as the effective path geometric component corresponding to the feature edge point. Repeat the process of acquiring the effective path geometric components corresponding to the feature edge points, acquire the effective path geometric components corresponding to each edge point in the target array region, and sort the acquired multiple effective path geometric components in descending order to obtain the path component sorting queue. The path component sorting queue is divided into several path component subsequences with the same interval span, and the obtained path component subsequences are named X1 component subsequence to Xa component subsequence respectively. It should be noted here that: In this application, X1, X2, X3...Xa in the X1 to Xa subsequences are respectively the flag symbols corresponding to the subsequences, and a is an integer greater than 0.

[0021] In the target region planar image, the edge points in the X1 component subsequence are marked as X1 edge points. The geometric center points of the areas covered by the sample array units in the target region planar image are obtained to obtain the geometric center points of the sample array. The lines connecting the X1 edge points and the geometric center points of the sample array are obtained to obtain multiple X1 edge lines. The spatial area covered by the obtained X1 edge lines in the target region planar image is set as the target region of the X1 sequence. Repeat the process of obtaining the target region of sequence X1, and obtain the target regions of sequence X2 to Xa respectively. Perform signal physical range coverage analysis on the edge points corresponding to the sample array area, and obtain the effective path geometric components corresponding to each edge point of the sample array area based on the analysis results. Acquire a planar image of the sample region, mark the edge points in the X1 component subsequence as X1 edge points, acquire the geometric center points of the area covered by the sample array unit in the planar image of the sample region, obtain the geometric center points of the sample array, acquire the lines connecting the X1 edge points and the geometric center points of the sample array, obtain multiple X1 edge lines, and set the spatial area covered by the acquired X1 edge lines in the planar image of the sample region as the X1 sequence sample region; Repeat the process of obtaining the X1 sequence sample region, and obtain the sequence sample regions corresponding to the edge points from X2 to Xa respectively, to obtain the X2 sequence sample region to the Xa sequence sample region; Obtain the area ratios of the target regions of sequences X1 to Xa to the target array region, respectively, to obtain the target area ratios of X1 to Xa. Also obtain the area ratios of the sample regions of sequences X1 to Xa to the sample array region, respectively, to obtain the sample area ratios of X1 to Xa. The channel path region deviation corresponding to the sample array unit is obtained by calculating the ratio of target area X1 to target area Xa and the ratio of sample area X1 to sample area Xa. The channel path region deviation corresponding to the sample array unit is calculated using the following formula: ; Where Qpd is the channel path region deviation degree corresponding to the sample array unit, Msbi is the target area ratio of Xi, Ysbi is the sample area ratio of Xi, and a is the quantity value corresponding to the path component subsequence. It should be noted here that: In this application, the target area ratio Xi involved herein can be any one of the target area ratios from X1 to Xa, and the sample area ratio Xi involved herein can be any one of the sample area ratios from X1 to Xa.

[0022] For different array units, the channel path region deviation between the sample channel field and the target channel field is obtained separately. A reference interval for the channel path region deviation is set. If the channel path region deviation is within the reference interval, the sample array area in the sample channel field is divided into a channel consistent area. If the channel path region deviation is not within the reference interval, the sample array area in the sample channel field is divided into a channel inconsistent area. It should be noted here that: In this application, the channel path area deviation corresponding to the historical channel consistency area is obtained, and multiple qualified path area deviations are obtained. The multiple qualified path area deviations are compared numerically, and the qualified path area deviation with the largest value is set as the upper limit of the channel path area deviation reference interval, and the qualified path area deviation with the smallest value is set as the lower limit of the channel path area deviation reference interval, thus obtaining the channel path area deviation reference interval. In this application, the channel consistency region referred to herein includes the case where the channel path region deviation is at the boundary of the channel path region deviation reference interval.

[0023] The number of consistent channel regions is counted as B1, and the number of inconsistent channel regions is counted as B2. B1 / (B1+B2) is calculated to obtain the overlap degree of the channel regions corresponding to the sample channel site and the target channel site. Repeat the process of obtaining the channel region overlap between the sample channel site and the target channel site. Obtain the channel region overlap between each historical channel site and the target channel site. Set the channel region overlap selection value. If the channel region overlap is greater than or equal to the channel region overlap selection value, select the corresponding historical channel site as the target matching site. If the channel region overlap is less than the channel region overlap selection value, select the corresponding historical channel site as a non-target matching site. Obtain the matching site selection data. It should be noted here that: In this application, when selecting historical target matching sites, the channel region overlap degree corresponding to the historical target matching site is obtained, and the values ​​of multiple obtained channel region overlap degrees are compared. The channel region overlap degree with the largest value is set as the channel region overlap degree selection value. It should be noted here that: The process of selecting the target matching site described above can be completed by training a deep learning model.

[0024] It should be noted here that: The above operations, combined with historical site data and signal unit-level environmental similarity analysis, significantly improve the reliability and applicability of channel prediction. First, matching is performed based on historical sites of the same type of antenna array, making full use of the verified hardware characteristics and channel response patterns to avoid prediction deviations caused by equipment differences. Secondly, by refining the environmental feature comparison down to the coverage area of ​​a single signal unit, multi-dimensional scene matching is achieved, making up for the neglect of local environmental differences by traditional macro-matching indicators; This two-layer matching mechanism not only inherits the empirical advantages of historical data, but also enhances the model's adaptability to complex scenarios through micro-environment calibration. This significantly improves the accuracy and robustness of channel characteristic prediction while reducing the actual measurement cost, providing a more reliable input basis for subsequent communication system optimization.

[0025] The site prediction module creates a weather signal impact model based on the matched site selection data, performs overall signal anomaly prediction for the target channel site based on the weather signal impact model, and obtains overall site signal prediction data based on the prediction results. Specifically as follows: Obtain matching site selection data, and obtain target matching sites based on the matching site selection data to obtain multiple target matching sites; Obtain the cumulative precipitation for each target matching site during the activity, obtain the cumulative precipitation for the activity period, and calculate the ratio of the cumulative precipitation for the activity period to the duration of the activity period to obtain the precipitation intensity for multiple sites. The average delay of the array signal corresponding to each target matching site during the activity is obtained, and the average delay of the signals of multiple sites is obtained. It should be noted here that: In this application, the average array signal delay referred to herein is the average value calculated by weighting the signal power of all signals from different paths when they arrive at the receiving antenna in a multipath propagation environment.

[0026] In this application, regarding weather factors affecting outdoor channel performance, precipitation typically has the most severe impact, primarily due to its direct interference with signal propagation through multiple mechanisms. First, water molecules in raindrops have a strong absorption effect on electromagnetic waves, especially in high-frequency bands (such as microwave and millimeter-wave communications), where signal energy is significantly consumed, leading to shortened propagation distance and signal strength attenuation. Secondly, dense raindrops induce significant scattering effects, causing signals to propagate along multiple paths, resulting in phase and amplitude distortion at the receiver and multipath interference. Furthermore, high humidity accompanied by rainfall further exacerbates atmospheric absorption and signal loss. In contrast, other weather factors such as fog and snow have lower dielectric constants and weaker attenuation effects on signals, while lightning, although capable of generating strong electromagnetic interference, is a transient event with a limited impact range. Therefore, precipitation, due to its persistence, diverse attenuation mechanisms, and wide-ranging impact on frequency bands, becomes the most significant weather factor affecting channel quality.

[0027] The acquired precipitation intensities at multiple sites are labeled as the first site precipitation intensity to the j-th site precipitation intensity in ascending order of their values. The average time delay of the site signals corresponding to the first site precipitation intensity to the j-th site precipitation intensity is set as the first site signal delay to the j-th site signal delay, respectively. It should be noted here that: In this application, j refers to the quantity value corresponding to the target matching site, and j is an integer greater than 0.

[0028] In practice, the following test data exists:

[0029] By performing polynomial fitting on the precipitation intensity from the first site to the precipitation intensity from the j-th site and the signal delay from the first site to the signal delay from the j-th site, a weather signal impact model is obtained. Specifically as follows: A polynomial fitting function is established from the precipitation intensity of the first site to the precipitation intensity of the j-th site and from the signal delay of the first site to the signal delay of the j-th site. The polynomial fitting function is as follows: ; Where y is the site signal delay, x is the site precipitation intensity, and a0 to a n The coefficients of the polynomial fitting function, where n is the order of the polynomial fitting function; It should be noted here that: In this application, site signal delay is the dependent variable and site precipitation intensity is the independent variable; Substituting the precipitation intensity from the first site to the precipitation intensity of the j-th site and the signal delay from the first site to the signal delay of the j-th site into the polynomial fitting function, the residual sum of squares (RSS) function of the site signal delay is calculated. The coefficients a0 to a10 of the polynomial fitting function in the residual sum of squares (RSS) function are respectively... n By taking partial derivatives, we obtain n unknowns a0 to a10. n The function expression, and n unknowns a0 to an By combining the function expressions, we obtain n sets containing a0 to a n The system of equations is obtained, and the system of equations is solved to obtain a0 to a n The specific value; From a0 to a n The specific numerical values ​​are substituted back into the polynomial fitting function to obtain the weather signal impact model; The predicted precipitation for the target channel site during the activity period is obtained by obtaining the weather forecast. The activity duration corresponding to the target channel site is obtained to obtain the activity duration of the target site. The ratio of the predicted precipitation to the activity duration of the target site is calculated to obtain the precipitation intensity of the target site. The precipitation intensity of the target site is input into the weather signal impact model to obtain the predicted signal delay value of the target channel site. A signal delay reference interval is set. If the predicted signal delay value of the site is within the signal delay reference interval, the target channel site is determined to be a signal analysis area. If the predicted signal delay value of the site is not within the signal delay reference interval, the target channel site is determined to be a signal abnormal site, and the overall signal prediction data of the site is obtained. It should be noted here that: In this application, the signal-qualified site referred to herein includes the case where the predicted site signal delay is within the boundary of the signal delay reference interval; Historical signal qualified sites are acquired, and the predicted signal delay value for each historical qualified site is obtained. The site with the largest predicted signal delay value is set as the upper limit of the signal delay reference interval. The lower limit of the signal delay reference interval involved here is 0, that is, there is no signal delay.

[0030] If the target channel site is a signal analysis area, then the channel direct bias analysis is performed on the target channel site, and the signal analysis area is further screened based on the analysis results to obtain the overall signal prediction data of the site. Specifically as follows: The feature array region is acquired in the target channel field, and a three-dimensional model of the feature array region is created to obtain the three-dimensional model of the feature array region. In the three-dimensional model of the feature array region, the active area of ​​the array service object is marked as the first array feature region, and the array signal transmission and reception area is marked as the second array feature region. The first array feature region is divided into several region pixels, and a first sample pixel is randomly selected from the obtained region pixels to divide the second array region into several region pixels. Please see Figure 3In the three-dimensional model of the feature array region, the plane where the second array feature region is located is marked as the angle reference plane. A line is drawn connecting each region pixel in the second array region to the first sample pixel to obtain multiple first signal direct connection mark lines. An auxiliary reference line is arbitrarily drawn in the angle reference plane. The angle between the first signal direct connection mark line and the auxiliary reference line at the first sample pixel is obtained numerically to obtain multiple channel transmit and receive angle values. The channel transmit / receive angle range corresponding to the sample array unit is obtained. If any channel transmit / receive angle value is within the channel transmit / receive angle range, the first sample pixel is marked as a directly connected channel pixel. If no such value exists, the first sample pixel is marked as a non-directly connected channel pixel. Repeat the process of classifying the first sample pixels, classify the pixels in each region of the first array feature region, and count the number of pixels in the direct channel to obtain the number of pixels in the direct channel. Calculate the ratio of the number of pixels in the direct channel to the number of pixels in the first array feature region to obtain the local proportion of the direct channel corresponding to the feature array region. Repeat the process of obtaining the local proportion of the direct channel corresponding to the feature array region, obtain the local proportion of the direct channel corresponding to each array region, and mark it as J1 direct channel local proportion to Jp direct channel local proportion. Obtain the area value of each array region, and mark it as J1 area value to Jp area value. It should be noted here that: In this application, J1, J2, J3...Jp in the area values ​​of region J1 to region Jp are array region marking symbols in the target channel field.

[0031] The area value of the target channel site is obtained by acquiring the area value of the target site. The direct connection degree of the array signal corresponding to the target channel site is obtained by calculating the ratio of the J1 direct channel local area to the Jp direct channel local area, the area value of J1 to the Jp area, and the area value of the target site. The direct connectivity of the array signal corresponding to the target channel site is calculated using the following formula: ; Where Zcl is the direct connection degree of the array signal corresponding to the target channel site, Sjyi is the local proportion of the direct connection channel in Ji, Zzbi is the area value of the Ji region, Sxd is the area value of the target site, and p is the number of array regions, where p is an integer greater than 0. It should be noted here that: The above formula objectively reflects the concentration of channel resources in spatial distribution through the combined effect of area and channel proportion, providing a quantitative basis for evaluating the performance of direct channel connections. Specifically, the formula uses the product of the local proportion of each direct channel (Sjyi) and the area (Zzbi) as the numerator, reflecting the dual influence of the distribution density (proportion) of the channel in the local area and the physical scale (area) of the area. Secondly, by normalizing the target site area (Sxd), the influence of site size differences on the results is eliminated, making the direct connection degree a relative indicator, which facilitates horizontal comparison between different sites. Finally, the contribution of all array regions (p) is integrated in a summation form, which not only preserves the physical meaning of multi-regional synergy but also ensures that the calculation results are decoupled from the granularity of the region division.

[0032] Set a preset range for array signal direct connection. If the array signal direct connection is within the preset range, the target channel area is classified as a signal qualified area. If the array signal direct connection is not within the preset range, the target channel area is classified as a signal abnormal area. It should be noted here that: In this application, the signal qualified field involved here includes the case where the array signal directness is within the boundary of the preset range of array signal directness; Historical channel sites that are judged to be qualified signal sites are acquired, and the array signal direct connection degree corresponding to each historical channel site is acquired. The array signal direct connection degree with the smallest value is set as the lower limit of the preset range of array signal direct connection degree. The upper limit of the preset range of array signal direct connection degree involved here is 100%, that is, there are no non-directly connected channel pixels in the target channel site.

[0033] The channel early warning module directly issues early warnings for sites with abnormal signals based on the overall site signal prediction data, and performs local signal obstruction analysis for sites with qualified signals, and issues local site early warnings based on the analysis results. Specifically as follows: Obtain overall site signal prediction data, and based on the overall site signal prediction data, identify sites with acceptable signal and sites with abnormal signal. If the target channel site is a channel abnormal site, a site channel abnormality warning will be issued directly. If the target channel site is a channel qualified site, a site local signal obstruction analysis will be performed on the target channel site, and a site local warning will be issued based on the analysis results. The different array regions contained in the target channel site are acquired, and a feature array region is arbitrarily selected from the acquired array regions. It should be noted here that: In this application, the deployment area of ​​the feature array region within the target channel site is specifically the working area where the feature array unit enhances the antenna signal; It should be noted here that: The above process constructs a multi-level channel analysis system of "overall prediction - hierarchical screening - local calibration", which significantly improves the precision and effectiveness of channel management; First, through holistic channel information prediction and early warning, abnormal channel locations can be quickly identified, enabling macro-optimization of network resources. Second, for qualified locations, direct channel obstruction analysis based on the lifting height of signal terminals is further implemented, which avoids redundant detection of qualified locations by the traditional "one-size-fits-all" evaluation mode. It can also accurately detect implicit channel attenuation caused by local environmental factors. This hierarchical processing mechanism not only improves the comprehensiveness of the prediction results—covering both the overall site conditions and local channel characteristics—but also provides precise guidance for site environment transformation through targeted early warnings, thereby effectively improving the service quality and reliability of the communication network while reducing operation and maintenance costs.

[0034] Signal occlusion analysis is performed on the feature array region, and the degree of direct signal occlusion corresponding to the feature array region is determined based on the analysis results. Specifically as follows: A three-dimensional model is created for the feature array region to obtain the feature region three-dimensional model. In the feature region three-dimensional model, the signal receiving and transmitting point layout plane corresponding to the feature array unit is set as the first region feature plane, and the ground area where the array service object is located is set as the second region feature plane. The first region feature plane is divided into several region pixels. A ray perpendicular to the second region feature plane is drawn through each region pixel to obtain multiple region pixel rays. Then, a sample region pixel ray is randomly selected from the multiple region pixel rays. The model region covered by the pixel ray of the sample region is divided into several ray pixels, and the vertical distance between each ray pixel and the ground is obtained to obtain the vertical distance of the pixel. Obtain the target matching site, select several array service objects in the target matching site, obtain the maximum lifting height of the terminal devices of each array service object in the target matching site, calculate the average of the obtained maximum lifting height of the terminal devices to obtain the first lifting height feature value, calculate the standard deviation of the obtained maximum lifting height of the terminal devices to obtain the second lifting height feature value, set the sum of the first lifting height feature value and the second lifting height feature value as the upper limit of the vertical distance target interval, obtain the minimum lifting height of each terminal device in the target matching site, calculate the average of the obtained minimum lifting height of the terminal devices to obtain the third lifting height feature value, calculate the standard deviation of the obtained minimum lifting height of the terminal devices to obtain the fourth lifting height feature value, set the difference between the third lifting height feature value and the fourth lifting height feature value as the lower limit of the vertical distance target interval, and obtain the vertical distance target interval; It should be noted here that: In this application, the array service target referred to herein is the audience; In this application, the terminal device referred to herein is specifically a wireless terminal communication device, including but not limited to mobile phones and tablets.

[0035] Pixels whose vertical distance is within the target vertical distance range are acquired to obtain multiple test pixels. The first region feature plane is divided into several planar pixels. Lines are drawn between each planar pixel and the test pixel to obtain multiple signal pseudo-direct connection paths. If there is no direct path among the acquired multiple signal pseudo-direct connection paths, the test pixel is classified as a non-directly oriented pixel. If any direct path exists among the acquired multiple signal pseudo-direct connection paths, the test pixel is classified as a directly oriented pixel. Repeat the process of classifying the pixel rays in the sample region, classify the pixel points to be tested contained in each region pixel ray, count the number of pixel points to be tested separated by the second region feature plane to obtain the first pixel feature value, and count the number of direct-ray pixel points separated by the second region feature plane to obtain the second pixel feature value. Calculate the difference between the feature value of the first pixel and the feature value of the second pixel, and calculate the ratio of the obtained difference to the feature value of the first pixel to obtain the direct signal occlusion degree corresponding to the feature array region; Repeatedly measure the direct signal occlusion degree corresponding to the feature array region, obtain the direct signal occlusion degree corresponding to each array region, and set a direct signal occlusion degree benchmark interval. If the direct signal occlusion degree is within the direct signal occlusion degree benchmark interval, it is determined that the signal occlusion of the corresponding array region is normal. If the direct signal occlusion degree is not within the direct signal occlusion degree benchmark interval, it is determined that the signal occlusion of the corresponding array region is abnormal, and a signal occlusion warning is issued. It should be noted here that: In this application, the normal state of signal obstruction in the array area referred to herein includes the case where the degree of direct signal obstruction is at the boundary of the direct signal obstruction reference interval. In this application, the lower limit of the direct occlusion reference interval is 0, that is, there is no signal occlusion. The historical characteristic array regions with normal signal occlusion in the array region are obtained to obtain multiple historical characteristic array regions. The direct signal occlusion degree corresponding to each historical characteristic array region is obtained, and the direct signal occlusion degree with the largest value is set as the upper limit of the direct occlusion reference interval.

[0036] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A deep learning-based system for predicting channel state information of large-scale antenna arrays, characterized in that, include: Data acquisition module: Performs array cell channel region overlap analysis on the target channel site and the historical channel site. Based on the analysis results, performs site similarity matching between the historical channel site and the target channel site to obtain matching site selection data. Site prediction module: Creates a weather signal impact model based on the matched site selection data, performs overall signal anomaly prediction for the target channel site based on the weather signal impact model, and obtains overall site signal prediction data based on the prediction results; Channel early warning module: It directly issues early warnings for sites with abnormal signals based on the overall signal prediction data of the site, and performs local signal obstruction analysis for sites with qualified signals, and issues local early warnings based on the analysis results; The data acquisition module is specifically used for: Obtain the target channel site and multiple historical channel sites, and arbitrarily select one sample channel site from the multiple historical channel sites. Array channel coverage analysis is performed on the sample channel site and the target channel site, and the overlap degree of the channel area corresponding to the sample channel site is obtained based on the analysis results. Obtain the channel area overlap degree corresponding to each historical channel site and the target channel site, set the channel area overlap degree selection value, if the channel area overlap degree is greater than or equal to the channel area overlap degree selection value, then select the corresponding historical channel site as the target matching site; if it is less than, then select the corresponding historical channel site as a non-target matching site, and obtain the matching site selection data. The array elements contained in the large-scale antenna array device are acquired, and a sample array element is randomly selected from the acquired array elements. The deployment area of ​​the sample array unit in the target channel field is set as the target array area, and the deployment area of ​​the sample array unit in the sample channel field is set as the sample array area. Perform signal physical range coverage analysis on the edge points corresponding to the target array area, and obtain the effective path geometric components corresponding to each edge point of the target array area based on the analysis results; Obtain the effective path geometric components corresponding to each edge point of the target array region, sort them in descending order to obtain the path component sorting queue, divide the path component sorting queue into several path component subsequences with the same interval span, and name the obtained path component subsequences from X1 component subsequence to Xa component subsequence respectively. In the subsequences X1 to Xa, X1, X2, X3...Xa are the corresponding markers of the subsequences, and a is an integer greater than 0. In the target region planar image, the edge points in the X1 component subsequence are marked as X1 edge points. The geometric center points of the sample array units in the target region planar image are obtained to obtain the geometric center points of the sample array. The lines connecting the X1 edge points and the geometric center points of the sample array are obtained to obtain multiple X1 edge lines. The spatial region covered by the obtained X1 edge lines in the target region planar image is set as the target region of the X1 sequence. Repeat the process of obtaining the target region of sequence X1, and obtain the target regions of sequence X2 to Xa respectively. Perform signal physical range coverage analysis on the edge points corresponding to the sample array area, and obtain the effective path geometric components corresponding to each edge point of the sample array area based on the analysis results. Acquire a planar image of the sample region, mark the edge points in the X1 component subsequence as X1 edge points, acquire the geometric center points of the area covered by the sample array unit in the planar image of the sample region, obtain the geometric center points of the sample array, acquire the lines connecting the X1 edge points and the geometric center points of the sample array, obtain multiple X1 edge lines, and set the spatial area covered by the acquired X1 edge lines in the planar image of the sample region as the X1 sequence sample region; Repeat the process of obtaining the X1 sequence sample region, and obtain the sequence sample regions corresponding to the edge points from X2 to Xa respectively, to obtain the X2 sequence sample region to the Xa sequence sample region; Obtain the area ratio of each sequence target region to the target array region to obtain multiple target area ratios; obtain the area ratio of each sequence sample region to the sample array region to obtain multiple sample area ratios. The channel path region deviation corresponding to the sample array unit is obtained by calculating the target area ratio and the sample area ratio. Set a channel path area deviation benchmark interval. If the channel path area deviation is within the channel path area deviation benchmark interval, the sample array area in the sample channel field is divided into a channel consistent area. If it is not within the benchmark interval, the sample array area in the sample channel field is divided into a channel inconsistent area. The number of consistent channel regions is counted as B1, and the number of inconsistent channel regions is counted as B2. The channel region overlap is then calculated. The site prediction module is specifically used for: Obtain matching site selection data, acquire target matching sites based on matching site selection data, obtain the ratio of cumulative precipitation to activity cycle duration for each target matching site during the activity, obtain multiple site precipitation intensities, and label them from smallest to largest as the first site precipitation intensity to the jth site precipitation intensity. The average time delay of the site signal corresponding to the precipitation intensity of the first site to the precipitation intensity of the j-th site is set as the time delay of the first site signal to the time delay of the j-th site signal, respectively. By performing polynomial fitting on the precipitation intensity from the first site to the precipitation intensity from the j-th site and the signal delay from the first site to the signal delay from the j-th site, a weather signal impact model is obtained. The precipitation intensity at the target site is obtained by the ratio of the predicted precipitation at the target site to the duration of the activity. The precipitation intensity of the target site is input into the weather signal impact model to obtain the predicted value of the site signal delay. A signal delay reference interval is set. If the predicted value of the site signal delay is within the signal delay reference interval, the target channel site is determined to be a signal assessment area. If it is not within the reference interval, the target channel site is determined to be a signal abnormality site. If the target channel site is a signal analysis area, then the channel direct bias analysis is performed on the target channel site. Based on the analysis results, the signal analysis area is further screened to obtain the overall signal prediction data of the site.

2. The deep learning-based channel state information prediction system for large-scale antenna arrays according to claim 1, characterized in that, Perform signal physical range coverage analysis on the edge points corresponding to the target array area, and obtain the effective path geometric components corresponding to each edge point of the target array area based on the analysis results, including: Acquire a planar image of the target region, mark the sample array units in the planar image of the target region, select a feature signal point, mark the edge of the target array region in the planar image of the target region, and select a feature edge point; In the planar image of the target region, the distance between the lines connecting the feature signal points and the feature edge points is obtained to get the projected distance of the model points. Create a 3D model of the target array region, and mark the vertical planes of feature points and edge points in the model. Collect the vertical distance between the vertical planes of feature points and edge points to obtain the longitudinal distance of the point model. In the 3D model of the target array region, a reference direct path angle is created to obtain the longitudinal distance Zjl of the point model and the projected distance Hjl of the model point. The actual distance of the field point is calculated, and the product of the actual distance of the field point and the cosine of the reference direct path angle is calculated to obtain the channel path geometric component between the feature edge point and the feature signal point. Obtain the channel path geometric components between the feature edge point and each signal point, and compare the values ​​of the obtained multiple channel path geometric components. Set the channel path geometric component with the smallest value as the effective path geometric component corresponding to the feature edge point.

3. The deep learning-based channel state information prediction system for large-scale antenna arrays according to claim 1, characterized in that, The channel early warning module is specifically used for: Obtain overall site signal prediction data, and based on the overall site signal prediction data, identify sites with qualified signals and sites with abnormal signals. If the target channel site is a channel abnormal site, a site channel abnormality warning will be issued directly. If the target channel site is a channel qualified site, a site local signal obstruction analysis will be performed on the target channel site, and a site local warning will be issued based on the analysis results. The different array regions contained in the target channel site are acquired, and a feature array region is arbitrarily selected from the acquired array regions. Signal occlusion analysis is performed on the feature array region, and the direct signal occlusion degree corresponding to the feature array region is obtained based on the analysis results; Obtain the direct signal obstruction degree corresponding to each array area and set the direct signal obstruction degree benchmark range. If the direct signal obstruction degree is within the direct signal obstruction degree benchmark range, it is determined that the signal obstruction of the corresponding array area is normal. If it is not within the benchmark range, it is determined that the signal obstruction of the corresponding array area is abnormal and a signal obstruction warning is issued.

4. The deep learning-based channel state information prediction system for large-scale antenna arrays according to claim 3, characterized in that, Signal occlusion analysis is performed on the characteristic array region, including: A three-dimensional model is created for the feature array region to obtain the feature region three-dimensional model. In the feature region three-dimensional model, the signal receiving and transmitting point layout plane corresponding to the feature array unit is set as the first region feature plane, and the ground area where the array service object is located is set as the second region feature plane. The first region feature plane is divided into several region pixels. A ray perpendicular to the second region feature plane is drawn through each region pixel to obtain multiple region pixel rays. Then, a sample region pixel ray is randomly selected from the multiple region pixel rays. The model region covered by the pixel ray of the sample region is divided into several ray pixels. The vertical distance between each ray pixel and the ground is obtained to obtain the vertical distance of the pixel.

5. The deep learning-based channel state information prediction system for large-scale antenna arrays according to claim 4, characterized in that, Signal occlusion analysis of the characteristic array region also includes: Several array service objects are selected in the target matching site. The maximum lifting height of the terminal equipment of the array service objects in the target matching site is numerically analyzed. The vertical distance target interval is obtained based on the analysis results. Pixels whose vertical distance is within the target vertical distance range are acquired to obtain multiple test pixels. The first region feature plane is divided into several planar pixels. Lines are drawn between each planar pixel and the test pixel to obtain multiple signal pseudo-direct connection paths. If there is no direct path, the test pixel is classified as a non-direct pixel. If there is a direct path, the test pixel is classified as a direct pixel. The types of the pixels to be tested contained in each region pixel ray are classified, and the number of pixels to be tested separated by the second region feature plane and the directly rayed pixels are counted to obtain the first pixel feature value and the second pixel feature value. Calculate the difference between the feature value of the first pixel and the feature value of the second pixel, and obtain the ratio of the obtained difference to the feature value of the first pixel to obtain the direct signal occlusion degree corresponding to the feature array region.

Images