CN-121995119-A - Spectrum map construction method combining spectrum mapping and data complement

CN121995119ACN 121995119 ACN121995119 ACN 121995119ACN-121995119-A

Abstract

The invention discloses a spectrum map construction method combining spectrum mapping and data complement, which solves the problem of mapping and complement process splitting in the prior art by introducing a Gaussian process to perform joint modeling on a spectrum space field, and realizes cooperative optimization of the two, then takes Gaussian process prediction uncertainty as a core to construct a sampling value evaluation index, guides monitoring equipment to perform self-adaptive sampling facing to a non-cooperative signal source coverage area so as to avoid data redundancy caused by blind sampling and improve data utilization rate, and finally enables a complement model to continuously reflect latest environment cognition and reversely guide subsequent sampling decision based on an online model updating mechanism of an extended data set, and finally remarkably improves the accuracy, stability and convergence efficiency of spectrum map construction under the condition of lower sampling cost.

Inventors

ZANG BO
ZHAO WEIHAO
LI BIN
LI KUN
WANG QIANNAN
Long Lulan
HE QIHONG
WEI HAITAO
LIU XUDONG

Assignees

西安电子科技大学

Dates

Publication Date: 20260508
Application Date: 20260210

Claims (8)

1. A method of spectrum map construction combining spectral mapping and data completion, the method comprising: Acquiring an initial sampling dataset in a spectral-spatial field; modeling the spectrum space field by using a Gaussian process to obtain a signal strength prediction mean value and a prediction variance at a current sampling point; constructing a sampling value evaluation index function based on the obtained prediction mean and the prediction variance, and obtaining a sampling position of a next target point; Planning a sampling path from a current position point to a next target point position, and sequentially collecting spectrum data along the planning path until reaching the target point position to obtain a newly added sampling data sample set; The acquired newly added sampling data sample set is integrated into the initial sampling data set to update the spectrum space model modeled by the Gaussian process; And repeating the modeling to updating the model process for iteration, and outputting the spectrum map.
2. The method for constructing a spectrum map by combining spectrum mapping and data complement according to claim 1, wherein the acquiring of the initial sampled data set specifically comprises: Let the three-dimensional spectrum space continuous region be At several sampling positions The measured signal intensity is The true electromagnetic radiation field is noted as Collecting N samples to form an initial data set The following formula: (1) in the formula, In order to measure the noise of the light, 。
3. The method for constructing a spectrum map by combining spectrum mapping and data complement according to claim 2, wherein the modeling specifically comprises: to unknown electromagnetic radiation field Modeling is a gaussian process, as follows: (2) in the formula, As a function of the a priori mean, Is a covariance kernel function; Wherein the covariance kernel function comprises a complex kernel function of shadow fading related terms and local smooth terms, and the covariance kernel function comprises the following steps of The complex covariance kernel function is expressed as: (3) Based on the initial data set By maximizing the log-marginal likelihood pair-fitting covariance kernel function superparameter and noise variance Estimating, initializing a Gaussian process regression model, and then collecting the existing extended data set of the t-th round Wherein After the t-th round of acquisition, any position to be predicted is acquired The predicted mean of (2) is as follows: (4) The prediction variance is as follows: (5)。
4. A method of spectral map construction in combination with spectral mapping and data completions as claimed in claim 3 wherein said evaluation index comprises an upper bound to confidence level And a spatial distance coverage term associated with the set of sampled points The evaluation function is specifically expressed as follows: (6) in the formula, Exploring weights for space; wherein, the confidence upper bound term The formula is as follows: (7) in the formula, For the normalized predicted expected value(s), For the normalized standard deviation of the predictions, Is a confidence coefficient; Space exploration weight The formula is as follows: (8) in the formula, For the initial model, the volume of the region of higher uncertainty, For the model t round iteration, the volume of the region with higher uncertainty is as follows: (9) (10) in the formula, To predict standard deviation Middle p quantile value.
5. The method for constructing a spectrum map by combining spectral mapping and data completion according to claim 4, wherein the obtaining of the sampling position of the next target point specifically includes selecting a point with highest average score in a neighborhood as a high-value target for the next acquisition by adopting a selection strategy of sampling points with neighborhood consistency : Neighborhood set defining location z , The following formula: (11) in the formula, Is the neighborhood radius; Calculating the neighborhood average score of each point in the region, and selecting the point with the maximum neighborhood average score as the target point of the next acquisition The following formula: (12)。
6. The method for constructing a spectrum map by combining spectrum mapping and data complement according to claim 5, wherein the path planning specifically comprises: Defining a discrete network reachable by the monitoring device: Setting the moving step length of the monitoring equipment as The set of reachable discrete network points is , And recording the point sequence of the sampling path to be planned as X, wherein the point sequence is represented by the following formula: (13) Recording each edge on the sampling path as an undirected line segment The set of past edges for the history is as follows: (14) The cost function is defined as follows: (15) in the formula, For the benefit of the information of the path, As a path loss term, In order to make the angle of rotation smooth, A launch direction penalty term, wherein, The formula is as follows: (16) Path loss term The formula is as follows: (17) corner smoothing term The formula is as follows: (18) Starting direction punishment item The formula is as follows: (19) in the formula, The end direction of the last section is the end direction; To sum up, the path planning process is modeled as an optimization problem as shown in the following equation: (20) in the formula, In order to be a safe distance constraint, To allow for a maximum scaling factor of the path length with respect to the start and end linear distance, Is the judgment of the situation that the projection sections are collinear and overlap; And solving the optimization problem to obtain a sampling path from the current position point to the position of the next target point.
7. The method for constructing a spectrum map by combining spectrum mapping and data complement as set forth in claim 6, wherein the updating process specifically includes: After t-round path exploration is executed, a new sampling data sample set is obtained The newly added sampling data sample set Incorporating historical data sets Forming an extended data set ; By expanding data sets Updating composite covariance kernel superparameter Using a maximum log-marginal likelihood MLE solution, the following formula: (21) Wherein: (22) Solving for The updated Gaussian process model parameters are formed and used as model parameters adopted in the calculation of the t+1-th round of prediction mean and prediction variance.
8. The method for constructing a spectrum map by combining spectrum mapping and data complement according to any one of claims 1 to 7, wherein the iterative process specifically comprises: global average uncertainty when prediction variance corresponds Below a preset threshold Or when the sampling resource occupation reaches a preset budget, terminating the sampling process and outputting a final spectrum map.

Description

Spectrum map construction method combining spectrum mapping and data complement Technical Field The invention relates to the technical field of spectrum map construction, in particular to a spectrum map construction method combining spectrum mapping and data complement. Background In recent years, with the rapid development of the emerging technologies such as 5G communication technology, internet of things, smart city, internet +' and the like, the demand of spectrum resources is in explosive growth, and relates to a plurality of fields of national economy and social life, meanwhile, the existing broad spectrum allocation mechanism is difficult to meet the continuously-rising frequency demand, so that the contradiction of spectrum resources is increasingly prominent, and the reasonable allocation and efficient utilization of spectrum resources are urgently realized through a fine management means. The electromagnetic spectrum map is used as an important technical means for multi-dimensionally describing an electromagnetic environment, can quantitatively describe and visually display the spatial distribution of electromagnetic energy and spectrum resources from multiple dimensions such as frequency, spatial position and field intensity and is combined with a geographic information system, and by constructing a high-precision spectrum map, the distribution situation of the spectrum resources can be comprehensively mastered, data support is provided for spectrum monitoring, interference analysis and resource scheduling, and the method has important significance for relieving shortage of the spectrum resources and improving the spectrum utilization efficiency. The prior art is as follows: CN202410446883 discloses a spectrum map mapping method for a complex urban environment, which starts from three-dimensional electromagnetic propagation modeling, discretizes a monitoring area into voxel grids and builds a propagation dictionary, determines the number of sampling points meeting reconstruction requirements and spatial distribution thereof through compressed sensing and eigenvalue threshold analysis, further adopts nearest neighbor strategy to plan unmanned aerial vehicle sampling tracks to complete data acquisition, and on the basis, firstly recovers sparse electromagnetic target signals, and then models and corrects shadow fading by utilizing a Gaussian process, thereby realizing the construction of a three-dimensional spectrum map. According to the method, the sampling positions and the number are mainly determined at one time based on offline threshold and dictionary analysis, the sampling process is difficult to dynamically adjust according to the model cognitive state, the completion process is mainly used as a post-processing correction means, the sampling decision is not guided reversely, so that the mapping and completion links are still relatively fractured, and the data utilization efficiency and the spectrum map reconstruction precision are difficult to be simultaneously considered under the condition of limited sampling cost. CN202411922547 discloses a method for constructing a spectrum coverage map based on preset acquisition paths and gridding modeling, which comprises the steps of defining an acquisition range according to base station parameters, planning the acquisition paths in advance, guiding monitoring equipment to traverse the acquired spectrum data along a set track, filling non-areas through gridding processing and a distance weighting/propagation model, and thus constructing the spectrum coverage map. The scheme has clear flow and simple realization, can complete large-scale spectrum mapping under lower system complexity, and has certain advantages in engineering realization and calculation cost. However, from the view of a sampling strategy, the sampling position of the method mainly depends on the traversal acquisition of a preset path, the geometric coverage is emphasized instead of the information value evaluation in the sampling process, and the difference and the dynamic adjustment of the contribution degree of different areas to the reconstruction of the spectrum map are difficult. In practical application, redundant sampling is easy to generate in a region with low information gain, and attention is insufficient on a high-value region, so that the overall data utilization rate is limited, the sampling efficiency is low, and the spectrum map construction performance is difficult to further improve under the constraint of limited sampling cost. The prior art has the following defects: 1. the mapping and data complement links are fractured, a joint modeling and collaborative optimization mechanism is lacked, and mapping efficiency and spectrum map reconstruction accuracy are difficult to be considered simultaneously under the constraint of limited sampling cost; 2. In the frequency spectrum mapping link, the existing sampling strategy mainly focuses on space geometric coverage, lac