CN114048783B - Cellular signal map construction method based on mobile group perception - Google Patents
Cellular signal map construction method based on mobile group perception Download PDFInfo
- Publication number
- CN114048783B CN114048783B CN202111365159.XA CN202111365159A CN114048783B CN 114048783 B CN114048783 B CN 114048783B CN 202111365159 A CN202111365159 A CN 202111365159A CN 114048783 B CN114048783 B CN 114048783B
- Authority
- CN
- China
- Prior art keywords
- signal
- features
- external
- tensor
- map
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000001413 cellular effect Effects 0.000 title claims abstract description 17
- 238000010276 construction Methods 0.000 title claims abstract description 15
- 230000008447 perception Effects 0.000 title claims abstract description 8
- 239000013598 vector Substances 0.000 claims abstract description 25
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 13
- 230000004927 fusion Effects 0.000 claims abstract description 12
- 238000000034 method Methods 0.000 claims description 31
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 6
- 230000007613 environmental effect Effects 0.000 claims description 5
- 238000013507 mapping Methods 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000005315 distribution function Methods 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims description 2
- 238000010586 diagram Methods 0.000 abstract description 23
- 238000011084 recovery Methods 0.000 abstract description 3
- 238000005259 measurement Methods 0.000 description 3
- 238000001228 spectrum Methods 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 2
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000007405 data analysis Methods 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000005562 fading Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention discloses a construction method for a cellular signal map based on mobile group perception, which stores space-time characteristics of signal data in a three-dimensional tensor, wherein three dimensions of the tensor respectively represent length, width and time dimensions. When a signal map is constructed, the potential factor characteristic of MCS signal data is acquired first, and then the internal characteristic of the MCS signal data is extracted. And then estimating missing signal values, and realizing signal recovery by utilizing Bayesian probability tensor decomposition on the basis of potential factor characteristic acquisition to obtain internal characteristics of the MCS data. Finally, a new multi-view feature fusion module based on learning is introduced, external features are embedded into a vector and combined with internal features, and then the combined features are input into a full connection layer to obtain signal features and construct a fine-granularity signal diagram.
Description
Technical Field
The invention relates to the field of signal map construction, in particular to a cellular signal map construction method based on mobile group perception.
Background
With the rapid development of the internet of things in the fields of smart cities and automatic driving services, the demands for wireless spectrum resources are increasing, and the wireless coverage environment is more complex. The complex and crowded wireless environment makes spectrum resources particularly valuable. To cater to this trend, signal diagrams are an effective indicator of the measurement of wireless environments, which can be applied to a variety of real world applications, including network operation, spectrum monitoring and location-based services. The signal map is a fingerprint database that marks the location information of a particular area. We estimate the location of the online fingerprint measurement by comparing with the fingerprint on the signal map. Furthermore, building and analyzing signal maps is a typical way to evaluate the availability of mobile broadband technologies (e.g. 4G/LTE, 5G) in different areas and compare different operators.
Since wireless signals are susceptible to various factors, such as automatic adjustment of transmission power, multipath fading, non-line-of-sight propagation, etc., it is difficult to maintain an up-to-date signal pattern efficiently and timely at the urban scale. Many studies on signal map construction rely on labor intensive and time consuming field surveys, given that the reference points are predefined, that must be performed periodically to maintain the latest signal map, resulting in extremely high costs. In order to reduce the costs incurred by intensive field investigation, there have also been some studies attempting to construct signal maps using model-based methods and data-driven methods.
The existing signal diagram construction method has two limitations: model-based approaches typically have strong assumptions about the distribution of signal patterns (e.g., ideal propagation space), which greatly limit their ability to model complex real-world wireless environments. Data driven methods require a large amount of measurement data that is typically collected by war drives and crowd sensing. But these data do not cover all areas of the network and do not closely reflect the actual experience of the user. The crowd sensing method does not process multi-level simultaneous interdependence relationship, and extracts spatial features in a limited local area. Furthermore, the fusion of signal internal features with external environmental features lacks a unified representation.
Disclosure of Invention
In order to solve the problems, the invention discloses a cellular signal map construction method based on mobile group perception, which comprises the following steps:
step 1, representing the space-time characteristics of the signal data as three dimensionsTensor S.epsilon.R I×J×K Wherein I, J represents the number of long and wide grids of the rectangular region; k represents the signal acquisition time span in the grid; s is S ijk Representing a signal value received by the mobile device from the base station;
step 2, extracting potential factor characteristics of MCS signal data, and calibrating signal values at a part of a BS coverage intersection area;
step 3, adopting Bayesian probability tensor decomposition to recover signal values of unobserved areas, namely estimating missing signals;
step 4, firstly constructing internal and external views; then a new multi-view feature fusion module based on learning is introduced, high-dimensional features are extracted and embedded into each view by utilizing two networks respectively, and then different views are spliced and fused by utilizing two full-connection layers, so that a fine and accurate signal diagram is generated.
The invention is further improved in that:
the signal data is more susceptible to environmental effects and exhibits fluctuating characteristics than other data sources (e.g., urban traffic data, noise data). We typically find an approximation by averaging over a given area. However, since characteristics (e.g., coverage, antenna transmit power, etc.) of each BS are different, it is not appropriate to directly average signals in some overlapping BS coverage areas, which may lead to signal drift and errors. More importantly, due to various unknown factors, we cannot effectively exploit the relationships between signals received by multiple BSs within the same grid. Therefore, when considering the tensor-based method, we need to extract the data satisfying the tensor low-rank feature, thereby constructing a tensor decomposition method suitable for signal data completion. That is, we need to scale the signal values in part at the intersection region before constructing the tensor and decomposition.
In the step (2), the latent factor feature extraction method is as follows:
step 201, describing the spatial correlation by using Euclidean distance, and then relating the geographic positions on the map, and using formula (1) to represent the distance correlation between BSs:
wherein sigma 2 Represents variance, Φ (i, j) represents the relationship of positions i, j, ||l i -l j || 2 Representing Euclidean distance, i.e
Step 202, using formula (2) to represent cosine similarity between two high-dimensional signal vectors, and performing another feature extraction:
wherein,h represents dimension, < >>Representing the signal vectors received from the H BSs in grid i.
The invention is further improved in that: by data analysis, we know that the signal data approximately obeys normal distribution;
in the step (3), the estimation of the missing signal value by adopting Bayesian probability tensor decomposition is carried out, and the specific method is as follows:
step 301. Assume that the noise term approximation of each observation term follows an independent Gaussian distribution, i.eThe row vectors of all factor matrices are further distributed a priori and assumed to be a multiple gaussian function, i.e +.>Wherein mu u ∈R R ，Λ u ∈R R×R And satisfies the formula (3) of probabilityThe distribution function is formula (4):
(μ u ，Λ u )～Gaussian-Wishart(μ 0 ，β 0 ，W 0 ，v 0 ) (3)
p(μ u ，Λ u |Θ)＝N(μ u |μ 0 ，(β 0 Λ u ) -1 )×Wishart(Λ u |W 0 ，v 0 )(4)
step 302. Since the accuracy of the signal data is unknown and it cannot be obtained completely by the inverse of the variance of all observations, in order to improve the robustness of the model, we introduce a flexible conjugate Gamma prior on the accuracy parameter τ, and express its prior function distribution by equation (5):
wherein the parameters U, V, W, τ and the super parameter μ u ，Λ u ，μ v ，Λ v ，μ w ，Λ w ，a 0 ，b 0 Solving by using Gibbs sampling;
step 303, factor matrix U, V, W may be passed through U i ，v i ，w i The solution is carried out, and the calculation method is as follows:
step 30301, obtaining a likelihood function of each observation term as shown in a formula (6) because the noise term of each observation term approximately obeys independent Gaussian distribution:
the formula is expressed as Hadamard product.
Step 30302, the row vector of the factor matrix is subjected to multi-element Gaussian distribution, and the likelihood function in step 30401 is combined to obtain a posterior distribution, wherein the posterior distribution is expressed as shown in a formula (7):
step 30303 obtaining the posterior distribution according to the step 30402 by using the formulas (8) and (9)
Step 304. Similarly, u is obtained by the method described above i ，v i ，w i Finally, after the Gibbs sampling algorithm reaches a stable state, all missing values can be estimated through Monte Carlo approximation, and then the estimation of missing signal data is realized.
The invention is further improved in that: in the step (4), the internal attempt and the external attempt are firstly carried out, and the specific method is as follows:
step 40101, taking the recovered ground signal tensor distribution and the characteristics thereof obtained in the step 3 as an internal view;
step 40102. The collected information data, including external characteristic signals of population, road network and POI, are mapped to tensors as external attempts.
5. The method for building a cellular signal map based on mobile group awareness according to claim 4, wherein in step (4), high-dimensional features are extracted and embedded for each view by using two networks, respectively, and the specific method is as follows:
40201 dividing tensor into matrix according to time dimension according to update period of signal to form coarse-granularity signal diagramIntroducing an average pooling layer to integrate the historical signal diagramExtracting features to obtain fine granularity signal diagram->
Step 40202, for the external view, a new context-aware neural network is proposed, embedding external features; population, road network and POI distribution characteristics, and representing them by embedded vectors; specifically, we first collect the external features and divide them by the same geographic space as the tensor. Then, in order to integrate the different types of background features and enhance the expression capability of the features, we embed the external feature sets of each small grid region to obtain an embedded vector of the background external features;
step 40203, mapping the fine-grained signal diagram obtained in step 40201 with the external feature vector obtained in step 40202, and processing by using a convolution layer to enhance reconstruction of external information;
the invention is further improved in that: in the step (4), the two full-connection layers are utilized to splice and fuse different views, so as to generate a fine and accurate signal diagram, and the specific method is as follows:
step 40301. Inputting the internal features and the external features into two multi-perception layers respectively, and then fusing them with a connection layer to obtain fine and accurate signal mapping;
step 40302 by providing training pairs using Adam optimizerTo learn the proposed model and to calculate the pixel-wise mean square error loss of the back-propagation;
step 40303 introducing a mask matrix M εR (NI N J) representing the region of available data in the fine-grained signal map, defining a loss functionTraining the fusion network to fuse the signal features with the external features and constructing a fine-granularity signal diagram based on the tensor recovered coarse-granularity signal diagram
The invention has the beneficial effects that:
two challenges are resolved, including: (1) missing and unreliable MCS data problems; (2) space-time uncertainty of signal propagation. The method for reconstructing the cellular signal diagram captures the space-time characteristics of signals between cellular base stations and in the cellular base stations at the same time, carries out missing signal recovery by using Bayesian tensor decomposition, and constructs a large-area fine-grained signal diagram. In addition, the method develops a context-aware multi-view fusion network to fully utilize external information and improve the construction precision of the signal diagram.
Drawings
Fig. 1: a flow chart of a cellular signal map construction method based on mobile group perception;
fig. 2: building a system frame diagram by using a cellular signal map;
fig. 3: a Bayesian tensor decomposition generating process;
fig. 4: a multi-view converged network structure;
fig. 5: a mobile group perception-based cellular signal map construction method flow chart.
Detailed Description
The present invention is further illustrated in the following drawings and detailed description, which are to be understood as being merely illustrative of the invention and not limiting the scope of the invention. It should be noted that the words "front", "rear", "left", "right", "upper" and "lower" used in the following description refer to directions in the drawings, and the words "inner" and "outer" refer to directions toward or away from, respectively, the geometric center of a particular component.
As shown in fig. 1-5, the present embodiment provides a cellular signal map construction method based on mobile group awareness, which includes the following steps:
step 1, representing the space-time characteristics of the signal data as a three-dimensional tensor S epsilon R I×J×K Wherein I, J represents the number of long and wide grids of the rectangular region; k represents the signal acquisition time span in the grid; s is S ijk Representing a signal value received by the mobile device from the base station;
step 2, extracting potential factor characteristics of MCS signal data, and calibrating signal values at a part of a BS coverage intersection area; the signal data is more susceptible to environmental effects and exhibits fluctuating characteristics than other data sources (e.g., urban traffic data, noise data). We typically find an approximation by averaging over a given area. However, since characteristics (e.g., coverage, antenna transmit power, etc.) of each BS are different, it is not appropriate to directly average signals in some overlapping BS coverage areas, which may lead to signal drift and errors. More importantly, due to various unknown factors, we cannot effectively exploit the relationships between signals received by multiple BSs within the same grid. Therefore, when considering the tensor-based method, we need to extract the data satisfying the tensor low-rank feature, thereby constructing a tensor decomposition method suitable for signal data completion. That is, we need to scale the signal values in part at the intersection region before constructing the tensor and decomposition. Inspired by BS knowledge extraction, we devised a suitable latent factor feature extraction method: in general, the closer the two locations, the higher the correlation of the two samples. We describe the spatial correlation by euclidean distance and then relate the geographic locations on the map. The distance correlation between BSs is defined as:
in addition, we also use cosine similarity between two high-dimensional signal vectors for another feature extraction, defined as:
step 3, adopting Bayesian probability tensor decomposition to recover signal values of unobserved areas, namely estimating missing signals;
by data analysis, we know that signal data approximately obeys normal distribution. Thus, we assume that the noise term of each observation term approximately follows an independent Gaussian distribution and is defined asWe further distribute the row vectors of all factor matrices a priori and assume that they obey a multivariate gaussian distribution +.>Super parameter mu u ∈R R ，Λ u ∈R R×R . This will enhance the robustness of the model and increase the convergence rate when sampling algorithms are used for model reasoning. Wherein the super parameter mu u ，Λ u Is defined as
(μ u ，Λ u )～Gaussian-Wishart(μ 0 ，β 0 ，W 0 ，v 0 )
p(μ u ，Λ u |Θ)＝N(μ u |μ 0 ，(β 0 Λ u ) -1 )×Wishart(Λ u |W o ，v 0 )
Since the accuracy of the signal data is unknown, and it cannot be captured entirely by the inverse of all observed value variances. In order to improve the robustness of the model, we introduce a flexible conjugate Gamma prior on the model:
fig. 2 is an overall diagram structure describing the bayesian tensor decomposition generation process described above. Wherein the parameters U, V, W, τ and the super parameter μ u ，Λ u ，μ v ，Λ v ，μ w ，Λ w ，a 0 ，b 0 Can be solved by gibbs sampling. Since each observation obeys an independent gaussian distribution, its likelihood can be written:
obtaining posterior distribution:
then, the following steps are carried out:
similarly, v j ，w k The method may be performed as described above. Finally, after the Gibbs sampling algorithm reaches a stable state, all missing values can be estimated through Monte Carlo approximation, and then the estimation of missing signal data is realized.
Step 4, firstly constructing internal and external views; then a new multi-view feature fusion module based on learning is introduced, high-dimensional features are extracted and embedded into each view by utilizing two networks respectively, and then different views are spliced and fused by utilizing two full-connection layers, so that a fine and accurate signal diagram is generated.
Internal and external view configuration
And after the tensor restores the missing signal, obtaining a coarse-granularity signal coverage map of the designated area. Although tensors have a good spatial structure to mine the spatio-temporal information, the recovered signals in the same grid may differ significantly at different times due to the dynamic environment. The signal map is considered to be constant over a certain update period. However, it is inaccurate to average only the historical signal patterns at different times to obtain a final signal pattern that is coarse-grained. Because the external features (population, POI, road network) in adjacent grids are similar, the signals received in the grids also have a certain similarity, which is helpful for constructing a fine-grained signal graph. Therefore, to improve the accuracy of the signal diagram, we further realize the fusion of the internal and external features of the multiple views, and we regard the recovered signal distribution tensor and its feature view as the a view. The collected information, including the demographics of population, road network and POIs, is mapped to tensors as B-view.
Multi-viewpoint neural fusion
The network fusing the different views to construct the final signal diagram is shown in fig. 3. The method comprises the steps of firstly extracting and embedding high-dimensional features from each view by using two networks, and then splicing and fusing different views by using two full-connection layers to generate a fine and accurate signal diagram.
In fig. 4, for view a, a complete signal space-time characteristic is represented in tensor form. Dividing tensors into matrices, i.e. coarse-grained signal maps, according to the update period of the signal mapIntroducing an average pooling layer, integrating the features extracted from the historical signal map to obtain a fine-grained signal map +.>For the B view we propose a new context aware neural network, embed the distribution features of external features (population, road network and POI) and represent them with embedded vectors. Specifically, we first collect the external features and divide them by the same geographic space as the tensor. Then, to integrate these different types of background features, we embed these external feature sets for each small grid region to get embedded vectors of the background external features, enhancing the expressive power of the features. Finally, we connect this with fine-grained signal mapping, and then process it with convolution layer, enhancing the reconstruction process of external information.
In order to integrate two parts of feature embedding, the internal features and the external features are firstly respectively input into two multi-perception layers, and then are integrated with a connecting layer to obtain fine and accurate signal mapping. We use Adam optimizer to provide training pairs by providing training pairsTo learn the proposed model and to calculate the pixel-wise mean square error loss of the back-propagation. We introduce a mask matrix M εR NI×NJ The areas where there is data available are represented in the fine-grained signal map. We define a loss function:
the fusion network solves the problem of space granularity division, enables the fusion of signal features and external features to be possible, and constructs a fine granularity signal diagram on the basis of a coarse granularity signal diagram of tensor recovery.
The technical solution of the present embodiment may be applied to the following embodiments:
example 1:
the method is used for constructing a high-quality and fine-grained signal map.
And according to the MCS space-time data information and in combination with external environment information (such as population, roads, POIs and BS positions), recovering the signal data is realized, so that a fine-grained signal map is constructed.
Example 2:
and updating and maintaining the cellular signal network in real time.
Because of environmental impact and population mobility, signal data has strong space-time dependence, so that the signal graph needs to be updated in real time. And constructing a signal diagram according to the current data information by using a cellular signal diagram construction system, and providing updating for a user.
Example 3:
for spatial localization, providing geographic information services.
And acquiring signal data by utilizing a cellular signal diagram construction system, and analyzing the distance correlation between the signal intensity and the base station, so that the signal source space positioning is realized, and corresponding geographic information service is provided for users.
This example to evaluate CSMC performance, we conducted extensive experimental and ablative studies on large-scale datasets acquired in the open sea above 200GB MCS signal recordings. The experimental results are shown in tables 1 and 2, and compared with the prior art, the signal estimation error performance of the method is improved by 29%.
TABLE 1 MAPE and RMSE results (City) for different methods at 30%, 60%, 90% sampling rate
TABLE 2 MAPE and RMSE results for different methods (suburban) at 30%, 60%, 90% sampling rate
The technical means disclosed by the scheme of the invention is not limited to the technical means disclosed by the embodiment, and also comprises the technical scheme formed by any combination of the technical features.
Claims (1)
1. The cellular signal map construction method based on mobile group perception is characterized by comprising the following steps of:
step 1, representing the space-time characteristics of the signal data as a three-dimensional tensorWherein I and J respectively represent the grid number of the length and the width of the rectangular area, K represents the signal acquisition time span in the grid, S ijk S represents the signal strength value received by the mobile device from the base station;
step 2, extracting potential factor characteristics of MCS signal data, and calibrating signal values at a part of a base station covering a crossing area;
in step 2, the latent factor feature extraction method is as follows:
step 201, describing the spatial correlation by using Euclidean distance, and then relating the geographic positions on the map, and using formula (1) to represent the distance correlation between the base stations:
wherein sigma 2 Represents variance, phi bs (i, j) represents the relationship of grids i and j,representing Euclidean distance, i.e
Step 202, using formula (2) to represent cosine similarity between two high-dimensional signal vectors, and extracting signal dimension characteristics:
wherein,h represents dimension, < >>Representing signal vectors received from H base stations in grid i;
step 3, adopting Bayesian probability tensor decomposition to recover signal values of unobserved areas, namely estimating missing signals;
in the step 3, the estimation of the missing signal value by adopting Bayesian probability tensor decomposition is carried out, and the specific method is as follows:
step 301. Assume that each observation term S ijk E S approximately obeys an independent Gaussian distribution, i.eFurther to the line vector U of the factor matrix U, V, W i ，v j ，w k A priori distribution is performed and,and all obey a multivariate gaussian distribution; the i-th row vector of the factor matrix U>Wherein mu u ∈R R ，∧ u ∈R R×R And satisfies the formula (3), (mu) u ，Λ u ) The probability distribution function of (2) is equation (4):
(μ u ，Λ u )～Gaussian-Wishart(μ 0 ，β 0 ，W 0 ，v 0 ) (3)
p(μ u ，Λ u |Θ)＝N(μ u |μ 0 ，(β 0 Λ u ) -1 )×Wishart(Λ u |W 0 ，v 0 ) (4)
step 302, introducing a flexible conjugate Gamma prior to the precision parameter tau, and expressing the precision parameter tau prior function distribution by using a formula (5):
wherein the parameters U, V, W, τ and the super parameter μ u ，∧ u ，μ v ，∧ v ，μ w ，∧ w ，a 0 ，b 0 Solving by using Gibbs sampling;
step 303, factor matrix U, V, W passes U i ，v j ，w k Solving, and the calculation method is as follows:
step 30301 due to each observation item S ijk Approximately obeying independent Gaussian distribution, likelihood functions can be obtained, and the expression is shown in a formula (6):
in the formula, +.;
step 30302, due to the row vector U of the factor matrix U i Obeying a multivariate gaussian distribution, in a joint step 30301Likelihood function of (2) by Bayes formula calculation to obtain row vector u i The expression is shown in formula (7):
step 30303, obtaining the posterior distribution according to the step 30302 by using the formulas (8) and (9)
Step 304, solving the formula (3) - (9) to obtain a row vector V of the factor matrix V, W j ,w k Finally, after the Gibbs sampling algorithm reaches a stable state, all missing values can be estimated through Monte Carlo approximation, so that the estimation of missing signal data is realized;
step 4, firstly, constructing an internal view and an external view; then, extracting high-dimensional features from each view by using two embedded networks and carrying out vector representation; finally, splicing and fusing different view vectors by using two full-connection layers to finally generate a refined signal map;
step 401, respectively constructing an internal signal view and an external environment view, and specifically, the method comprises the following steps:
step 40101, taking the recovered signal tensor distribution obtained in the step (3) and the characteristics thereof as an internal view;
step 40102. Map the collected information data, including external feature signals of population, road network and points of interest, into tensors as external views;
step 402, extracting high-dimensional features from the internal and external views by using two embedded networks and performing vector representation, wherein the specific method comprises the following steps:
40201 dividing the recovered tensor signal distribution obtained in step 3 into matrix according to time dimension and update period of the signal to form coarse-granularity signal mapIntroducing an average pooling layer, integrating the features extracted from the historical signal map to obtain a fine-grained signal map +.>
Step 40202, for the external view, a context-aware based external feature embedding representation is proposed; the population, road network and the distribution characteristics of interest points are mapped into tensors and expressed by embedded vectors; specifically, firstly, collecting external features, and dividing an external view according to the geographic space which is the same as tensor S, namely, geographic spaces with length and width of I and J grids respectively; then, in order to integrate the different types of environmental features, embedding an external feature set of each small grid area to obtain an embedded vector of the external features;
step 40203, mapping the fine-grained signal map obtained in step 40201 with the external environment feature vector obtained in step 40202, and processing by using a convolution layer to enhance reconstruction of external information;
step 403, splicing and fusing different views by using two full-connection layers to generate a fine and accurate signal map, wherein the specific method is as follows:
step 40301, respectively inputting the internal signal features and the external environment features into two multi-layer sensing layers, then splicing and fusing output results, and obtaining a fine and accurate signal map through a full-connection layer;
step 40302 by providing training pairs using Adam optimizerTo learn the proposed multi-view fusion model and calculateBack-propagating pixel-wise mean square error loss;
step 40303 introducing a mask matrix M εR (NI N J) representing regions of observable data in the fine-grained signal plot, defining a loss functionTraining the multi-view fusion model, fusing the internal time-varying signal characteristics output by the step 40201 with the external environment characteristics output by the step 40202, and recovering a coarse-granularity signal map in tensor->On the basis of the above, a fine-grained accurate signal map is constructed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111365159.XA CN114048783B (en) | 2021-11-17 | 2021-11-17 | Cellular signal map construction method based on mobile group perception |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111365159.XA CN114048783B (en) | 2021-11-17 | 2021-11-17 | Cellular signal map construction method based on mobile group perception |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114048783A CN114048783A (en) | 2022-02-15 |
| CN114048783B true CN114048783B (en) | 2024-04-16 |
Family
ID=80209964
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111365159.XA Active CN114048783B (en) | 2021-11-17 | 2021-11-17 | Cellular signal map construction method based on mobile group perception |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114048783B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103745489A (en) * | 2014-01-07 | 2014-04-23 | 电子科技大学 | Method for constructing base station signal field intensity map based on compressed sensing |
| CN110223509A (en) * | 2019-04-19 | 2019-09-10 | 中山大学 | A kind of missing traffic data restorative procedure enhancing tensor based on Bayes |
| CN111757258A (en) * | 2020-07-06 | 2020-10-09 | 江南大学 | Self-adaptive positioning fingerprint database construction method under complex indoor signal environment |
| CN112532439A (en) * | 2020-11-24 | 2021-03-19 | 山东科技大学 | Network flow prediction method based on attention multi-component space-time cross-domain neural network model |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7359718B2 (en) * | 2004-04-30 | 2008-04-15 | Hong Kong Applied Science And Technology Research Institute Co., Ltd. | Location determination and location tracking in wireless networks |
-
2021
- 2021-11-17 CN CN202111365159.XA patent/CN114048783B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103745489A (en) * | 2014-01-07 | 2014-04-23 | 电子科技大学 | Method for constructing base station signal field intensity map based on compressed sensing |
| CN110223509A (en) * | 2019-04-19 | 2019-09-10 | 中山大学 | A kind of missing traffic data restorative procedure enhancing tensor based on Bayes |
| CN111757258A (en) * | 2020-07-06 | 2020-10-09 | 江南大学 | Self-adaptive positioning fingerprint database construction method under complex indoor signal environment |
| CN112532439A (en) * | 2020-11-24 | 2021-03-19 | 山东科技大学 | Network flow prediction method based on attention multi-component space-time cross-domain neural network model |
Non-Patent Citations (2)
| Title |
|---|
| 基于张量分解的大规模多输入多输出天线预编码;陈文娟;周小平;王家南;李莉;杨哲;;上海师范大学学报(自然科学版)(02);全文 * |
| 基于贝叶斯张量分解的改进QoS预测;谭朝文;宗容;席先强;武浩;;计算机工程与应用(20);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114048783A (en) | 2022-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ochilov et al. | Operational SAR sea-ice image classification | |
| Casali et al. | A topological analysis of growth in the Zurich road network | |
| Sarkar et al. | Modeling spatial determinants of urban expansion of Siliguri a metropolitan city of India using logistic regression | |
| CN105911518A (en) | Robot positioning method | |
| Kirthika et al. | Automated road network extraction using artificial neural network | |
| Han et al. | Groundwater level prediction using a SOM-aided stepwise cluster inference model | |
| CN109154938B (en) | Classifying entities in a digital graph using discrete non-trace location data | |
| CN111539453B (en) | Global ionized layer electron total content prediction method based on deep cycle neural network | |
| CN112911626B (en) | Wireless network flow prediction method based on multi-graph convolution | |
| Lee et al. | High-resolution poverty maps in sub-saharan africa | |
| Li et al. | Bayesian Markov chain random field cosimulation for improving land cover classification accuracy | |
| Tang et al. | Dynamic origin‐destination flow estimation using automatic vehicle identification data: A 3D convolutional neural network approach | |
| Xi et al. | Beyond the first law of geography: Learning representations of satellite imagery by leveraging point-of-interests | |
| Pandey et al. | Numerical modelling spatial patterns of urban growth in Chandigarh and surrounding region (India) using multi-agent systems | |
| Rubio et al. | Adaptive non-parametric identification of dense areas using cell phone records for urban analysis | |
| Cheng et al. | A lightweight ensemble spatiotemporal interpolation model for geospatial data | |
| Chen et al. | Graph-based radio fingerprint augmentation for deep-learning-based indoor localization | |
| CN114048783B (en) | Cellular signal map construction method based on mobile group perception | |
| CN113590971A (en) | Interest point recommendation method and system based on brain-like space-time perception characterization | |
| Zhang et al. | Dac-ml: domain adaptable continuous meta-learning for urban dynamics prediction | |
| Gkillas et al. | Federated learning for lidar super resolution on automotive scenes | |
| CN114818857A (en) | Deep snow fusion method | |
| Chen et al. | Bayesian subpixel mapping of hyperspectral imagery via discrete endmember variability mixture model and markov random field | |
| Harmassi et al. | Smart things: conditional random field based solution for context awareness at the IoT edge | |
| Gómez et al. | WiFi-based Localization for Fail-Aware Autonomous Driving in Urban Scenarios |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |