CN110381440A - The fingerprint indoor orientation method of joint RSS and CSI based on deep learning - Google Patents
The fingerprint indoor orientation method of joint RSS and CSI based on deep learning Download PDFInfo
- Publication number
- CN110381440A CN110381440A CN201910518903.1A CN201910518903A CN110381440A CN 110381440 A CN110381440 A CN 110381440A CN 201910518903 A CN201910518903 A CN 201910518903A CN 110381440 A CN110381440 A CN 110381440A
- Authority
- CN
- China
- Prior art keywords
- csi
- rss
- fingerprint
- joint
- deep
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S5/00—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations
- G01S5/02—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations using radio waves
- G01S5/0257—Hybrid positioning
-
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/02—Services making use of location information
- H04W4/021—Services related to particular areas, e.g. point of interest [POI] services, venue services or geofences
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/30—Services specially adapted for particular environments, situations or purposes
- H04W4/33—Services specially adapted for particular environments, situations or purposes for indoor environments, e.g. buildings
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Position Fixing By Use Of Radio Waves (AREA)
Abstract
The invention belongs to wirelessly communicate and indoor positioning technologies field, disclose the fingerprint indoor orientation method of joint RSS and CSI based on deep learning a kind of, off-line phase is collected and processing RSS and CSI information, and CSI of the training based on deep layer self-encoding encoder handles model, to the CSI coding at reference point, joint RSS and CSI coding building fingerprint base, divides sub fingerprint library according to AP selection algorithm;For the training of each sub fingerprint library, one is based on deep neural network location prediction model;The tuning on-line stage encodes the CSI of tested point according to the trained CSI processing model based on deep layer self-encoding encoder of off-line phase, calculates joint RSS and CSI fingerprint, based on deep neural network model prediction positioning.The present invention combines RSS and CSI information, the computation complexity and time-consuming for enriching finger print information, reducing location prediction model;Positioning accuracy is high, and scalability is strong, and transportable property is high.
Description
Technical field
The invention belongs to wirelessly communicate and indoor positioning technologies field more particularly to a kind of joint based on deep learning
The fingerprint indoor orientation method of RSS and CSI.
Background technique
Currently, the immediate prior art: with universal, the indoor positioning based on WLAN of wireless network and mobile device
Receive more and more attention, positioning accuracy can achieve that meter level is other, without disposing additional equipment, position it is at low cost, easily
It is promoted in the application in daily life.Indoor positioning technologies based on WLAN can be divided into two kinds: the positioning based on ranging is calculated
Method (Ranging-based Localization) and indoor algorithm (Fingerprint-based based on location fingerprint
Localization).Since indoor environment is complicated and changeable, wireless signal is in communication process due to shielding blocking of building etc.
Factor easily occurs reflection, refraction and scattering, generates multipath effect.Therefore, the indoor locating system mainstream based on WLAN uses
Location technology based on location fingerprint.And base can be divided into according to the difference of fingerprint classification based on WLAN fingerprint base location algorithm
In received signal strength (Received Signal Strength, RSS) fingerprint and it is based on channel state information (Channel
State Information, CSI) fingerprint indoor orientation method.
Wherein, it is to be very easy to acquisition the advantages of RSS, so that positioning system is easily realized, easy promotion and popularization.And RSS
The shortcomings that there are two main.Firstly, RSS is the coarseness of MAC sublayer (Media Access Control, MAC)
Wireless signal information is only capable of embodying time-domain information.More importantly RSS is changed over time greatly, and sensitive to multipath fading,
I.e. stability is poor, this will bring very big challenge for the predictive ability of location model (classifier or prediction model), and then lead
It causes positioning accuracy unstable and is difficult to be promoted.Indoor orientation method based on RSS fingerprint for example, RADAR system, substantially
It is a kind of based on K nearest neighbor algorithm (K nearest neighbor, KNN): according to the RSS data of user's test point on line in RSS
In the feature space of fingerprint base, with access point (Access Point, AP) be characterized K similarity of search (in the location algorithm,
Using the Euclidean distance between RSS fingerprint vector between position similarity, the RSS fingerprint vector Euclidean distance of even two positions
It is about small, then illustrate that its distance is about close) highest reference point, the mean place for calculating this K reference point is the object of user's test point
Manage position.
Although CSI is a kind of fine-grained wireless signal information of physical layer (Physical Layer) compared with RSS, can
To provide time-domain and frequency-domain information or amplitude and phase information abundant.But CSI is a kind of high dimensional data, and different
There are redundancies between subcarrier.If directly positioned using CSI data, the computation complexity that will lead to location model is high and very
Difficult accurate positioning.Indoor orientation method based on CSI fingerprint for example, fine granularity fingerprint indoor locating system (Fine-
Grained Indoor Fingerprinting System, FIFS) main thought be to utilize the CSI amplitude on three antennas
Weighted average establish fingerprint base as location fingerprint feature, and realize location algorithm based on probability.FIFS algorithm is simple,
And complexity is low, but positioning accuracy is limited.In order to which the CSI data bulk redundancy information coped between different sub-carrier is with noise
The problem of location algorithm is brought, some researchs are first to CSI data prediction.For example, utilizing Principal Component Analysis
(Principal Component Analysis, PCA) and linear discriminant analysis (Linear Discriminant
Analysis, LDA) to both typical dimension-reduction algorithms to CSI data prediction, further, training classifier to position, such as
Naive Bayesian, support vector machines (Support Vector Machine, SVM), random forest (Random Forest, RF).
DeepFi is a kind of indoor orientation method based on deep learning, and off-line phase utilizes the CSI width for 90 subcarriers being collected into
Information is spent as fingerprint, and successively stack is trained to be limited Boltzmann machine (Restricted Boltzmann using greedy algorithm
Machine, RBM);When tuning on-line, CSI amplitude fingerprint first at collection test point, by the stack of each reference point by
It limits Boltzmann machine and calculates reconstructed error, and calculate the probability of test point Yu the reference point with this, finally complete to test point
Positioning.DeepFi is substantially that one stack of each reference point training of root is limited Boltzmann machine, when the larger reference point of localizing environment
When more, computation complexity and the training time of DeepFi algorithm will obviously increase sharply.On the other hand, the master of DeepFi localization method
It is intended to consider that CSI's at single reference point is abstract, at the same time, has ignored the connection between different reference points.
In conclusion problem of the existing technology is: due to the coarseness and unstable feature of RSS, being referred to based on RSS
Line positioning accuracy and stability is not high;And CSI information is the challenge of positioning accuracy band there are redundancy and noise problem.
Solve the difficulty of above-mentioned technical problem:
It is limited and steady all to there is positioning accuracy in the fingerprint positioning method based on RSS or the fingerprint positioning method based on CSI
Qualitative not high problem.The advantages of RSS, is very easy acquisition, this realizes positioning system easily, easy promotion and popularization.And
The shortcomings that there are two RSS is main.First, RSS is the thick of MAC sublayer (Media Access Control, MAC)
Granularity wireless signal information is only capable of embodying time-domain information.Second, RSS is changed over time greatly, and sensitive to multipath fading, i.e., surely
Qualitative difference, this will bring very big challenge for the predictive ability of location model (classifier or prediction model), and then cause to determine
Position precision be difficult promoted and it is unstable.Although CSI is a kind of fine-grained wireless signal information of physical layer, can provide abundant
Time-domain and frequency-domain information or amplitude and phase information.But CSI is a kind of high dimensional data, and is existed between different sub-carrier
Redundancy.If directly positioned using CSI data, it will lead to the computation complexity height of location model and be difficult accurate positioning.
Solve the meaning of above-mentioned technical problem:
The fingerprint indoor orientation method of the joint RSS and CSI based on deep learning proposed in the present invention, it is contemplated that different
Connection between reference point obtains the fingerprint based on CSI abstract characteristics;Joint RSS and CSI enriches location fingerprint feature;Based on depth
Degree learns the Nonlinear Mapping relationship of powerful learning ability study location fingerprint and physical location complexity, and then improves polarization
Energy and stability.
Summary of the invention
In view of the problems of the existing technology, the finger for the joint RSS and CSI that the present invention provides a kind of based on deep learning
Line indoor orientation method.
The invention is realized in this way a kind of fingerprint indoor orientation method of the joint RSS and CSI based on deep learning,
The fingerprint indoor orientation method of the joint RSS and CSI based on deep learning includes off-line phase and tuning on-line stage;
Off-line phase collects and handles RSS and CSI information, and CSI of the training based on deep layer self-encoding encoder handles model, right
CSI coding at reference point, joint RSS and CSI coding building fingerprint base, divide sub fingerprint library according to AP selection algorithm;For
Each the training of sub fingerprint library one is based on deep neural network location prediction model;
The tuning on-line stage is according to the trained CSI processing model based on deep layer self-encoding encoder of off-line phase to tested point
CSI coding, calculate joint RSS and CSI fingerprint, positioned based on deep neural network model prediction.
Further, the off-line phase of the fingerprint indoor orientation method of the joint RSS and CSI based on deep learning includes
Following steps:
Step 1, data collection and processing;
Step 2 designs the structure of deep layer autocoder model, including its according to the CSI amplitude information that processing obtains
Depth, neuron number and activation primitive, and the CSI based on deep layer self-encoding encoder handles mould accordingly for training as unit of AP
Type;
Step 3 encodes the CSI data at reference point according to trained encoder, and CSI coding is believed with RSS
New fingerprint base is established in the combination of breath;
Step 4, location prediction model of the training based on deep-neural-network, planned network structure and activation primitive, with fixed
Position error is target, specially minimizes the mean square error of normalization the prediction coordinate and true normalized coordinate of output;With phase
The fingerprint vector in the sub fingerprint library answered is input, and true normalized coordinate is output, and normalization physical location coordinate is output, instruction
Practice the location prediction model based on deep neural network.
Further, the data collection of the step 1 is specifically included with processing:
(1) indoors in positioning scene, coordinate system is established, reference point and test point is divided, receives at each position respectively
Collection data simultaneously record its relative coordinate;It for the data file of each reference point, is handled, is used using MATLAB software
Read_bf_file function in CSI Tool reads data;
(2) it arranges with processing data, the get_total_rss and get_scaled_csi letter being utilized respectively in CSI Tool
Number obtains and separates RSS and CSI information.
Further, the tuning on-line stage of the fingerprint indoor orientation method of the joint RSS and CSI based on deep learning
The following steps are included:
Step 1 handles the data of tested point, according to data processing module, handles the data of tested point, obtains
CSI amplitude information and RSS information;
Step 2 calculates the coding of CSI at tested point, is input with the CSI amplitude information of tested point, according to off-line phase
In encoder section in trained deep layer autocoder model, obtain its corresponding CSI encoded information;
Step 3 generates the joint RSS and CSI fingerprint of tested point, according to the CSI encoded information and RSS information of tested point,
And according to actual location scene, corresponding AP selection algorithm is added, the connection for generating tested point according to the step of off-line phase four
Close fingerprint vector;
Step 4 obtains the prediction result of tested point, using the fingerprint vector of the joint RSS and CSI of generation as based on deep
The input for spending the location prediction model of neural network exports to predict coordinate.
The fingerprint indoor positioning for combining RSS and CSI based on deep learning that another object of the present invention is to provide a kind of
The fingerprint indoor locating system of method, the joint RSS and CSI based on deep learning includes:
Data collection and processing module collect RSS and CSI information, handle mould by the CSI based on deep layer autocoder
Type carries out dimensionality reduction to CSI data and feature extraction is handled;
Fingerprint combination module, for being encoded by RSS and by the CSI of the CSI processing model based on deep layer autocoder
Combination, establishes fingerprint base;
Location prediction module, the data for being collected according to tested point predict coordinate, to combine the finger of RSS Yu CSI information
Line library data set is input, and by the location prediction model based on deep neural network, obtained output is to predict coordinate.
Another object of the present invention is to provide the fingerprints of the joint RSS and CSI described in application based on deep learning a kind of
The wireless communication system of indoor orientation method.
Another object of the present invention is to provide the fingerprints of the joint RSS and CSI described in application based on deep learning a kind of
The indoor locating system of indoor orientation method.
In conclusion advantages of the present invention and good effect are as follows: the influence due to factors such as indoor environments to RSS is very big,
So that the robustness based on RSS location algorithm is weaker, and this coarse grain information of RSS for positioning accuracy raising band very
Big challenge.CSI information can provide richer information, but CSI is positioning accuracy there are redundancy and noise problem
With challenge.To further increase positioning accuracy as target, the present invention is first of all for abundant finger print information, reduction location prediction mould
The computation complexity and time-consuming of type, have collected a large amount of RSS and CSI information, and the CSI based on deep layer self-encoding encoder by proposing
The new feature space mapping of characteristic extracting module acquisition CSI data will be located at.Then, combine RSS information and the coding of CSI is built
Vertical fingerprint base.Finally, devising the location prediction method based on deep-neural-network, positioning accuracy is high, and scalability is strong, can move
Shifting property is high.
As shown in Fig. 3 and table 1, compared with KNN, DeepFi, FIFS, svm+lda, rf+lda, it is proposed by the present invention to be based on
Fingerprint indoor orientation method (referred to as DADN) CDF curve of the joint RSS and CSI of deep learning, average localization error, intermediate value
Error, the contrast verifications of these experimental results of position error standard deviation DADN algorithm are able to ascend positioning accuracy and stability:
On the one hand, DADN algorithm joint CSI and RSS establishes fingerprint base, enriches position feature, has apparent gain to positioning accuracy.Separately
On the one hand, the DADN algorithm based on deep learning is for complicated between study these radio signal characteristics of CSI and RSS and position coordinates
Mapping ability it is stronger.
Detailed description of the invention
Fig. 1 is the fingerprint indoor orientation method stream of the joint RSS and CSI provided in an embodiment of the present invention based on deep learning
Cheng Tu.
Fig. 2 is that the fingerprint indoor orientation method of the joint RSS and CSI provided in an embodiment of the present invention based on deep learning is real
Existing flow chart.
Fig. 3 is the experimental result schematic diagram of positioning accuracy performance provided in an embodiment of the present invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to embodiments, to the present invention
It is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to
Limit the present invention.
It is not high with stability based on RSS fingerprint location precision for the coarseness and unstable feature due to RSS, it deposits
The challenge that has been positioning accuracy band of redundancy and noise problem the problem of.The present invention consider first joint RSS and CSI information with
Abundant fingerprint characteristic excavates fingerprint characteristic based on deep learning and realizes high-precision indoor orientation method.
Technical solution of the present invention is explained in detail with reference to the accompanying drawing.
As shown in Figure 1, the fingerprint indoor positioning of the joint RSS and CSI provided in an embodiment of the present invention based on deep learning
Method the following steps are included:
S101: data collection and processing module collect RSS and CSI information, at the CSI based on deep layer self-encoding encoder
Reason model carries out abstract and feature extraction to CSI, obtains CSI coding;
S102: joint RSS and CSI coding building fingerprint base, to enrich location fingerprint feature;
S103: the location prediction of the location prediction model prediction coordinate based on deep neural network.
The fingerprint indoor locating system of joint RSS and CSI provided in an embodiment of the present invention based on deep learning includes: number
According to collection and processing module, fingerprint combination module and location prediction module.
Data collection and processing module collect RSS and CSI information, handle mould by the CSI based on deep layer autocoder
Type carries out dimensionality reduction to CSI data and feature extraction is handled;
Fingerprint combination module, for being encoded by RSS and by the CSI of the CSI processing model based on deep layer autocoder
Combination, establishes fingerprint base;
Location prediction module, the data for being collected according to tested point predict coordinate, to combine the finger of RSS Yu CSI information
Line library data set is input, and by the location prediction model based on deep neural network, obtained output is to predict coordinate.
Technical scheme of the present invention will be further described with reference to the accompanying drawing.
As shown in Fig. 2, the fingerprint indoor positioning of the joint RSS and CSI provided in an embodiment of the present invention based on deep learning
Including off-line phase and tuning on-line stage.
Off-line phase the following steps are included:
Step 1, data collection and processing:
(1a) in positioning scene, establishes coordinate system indoors, divides reference point and test point, receives at each position respectively
Collection data simultaneously record its relative coordinate.For preferably trained deep neural network model, need to collect a large amount of data, for
It in the data file of each reference point, is handled using MATLAB software, uses the read_bf_file letter in CSI Tool
Number reads data.
(1b) is arranged and processing data:
Further, the get_total_rss and get_scaled_csi function being utilized respectively in CSI Tool are obtained and are divided
From RSS and CSI information.
Step 2, the CSI based on deep layer self-encoding encoder handle model:
According to the CSI amplitude information handled in step 1, the structure of deep layer autocoder model, including its are designed
The selection of depth, neuron number and activation primitive.If deep layer autocoder model is regarded as a black box, function
It can be the potential data characteristics for exploring CSI data set, be converted with the subcarrier dimensional space of its different antennae to embodying as far as possible
Dimensionality reduction and feature extraction functions to CSI data are realized in the new space of CSI data characteristics.Deep layer autocoder model it is defeated
Enter be an AP CSI data, export it is identical with input, target be by the labyrinths of intermediate hidden layers with it is thousands of on
Ten thousand parameter minimizes the error of input with output, embodies the CSI data inner link on corresponding AP, completes " encoder-
The function of decoder ".
Step 3, fingerprint combination module construct fingerprint base:
The CSI data at reference point are encoded according to " encoder " trained in step 2.Further, it is possible to plus
Enter AP selection algorithm etc. to be further processed data.Finally, new fingerprint base is established into the combination of CSI coding and RSS information,
To enrich fingerprint characteristic.
Step 4, the location prediction based on deep-neural-network:
Planned network structure and activation primitive, the mean square error of prediction coordinate and actual coordinate by minimizing output,
I.e. position error is target, and the fingerprint vector with joint RSS and CSI coding in fingerprint base is input, normalizes physical location
Coordinate is output, training deep neural network model.
The tuning on-line stage of the invention includes the following steps:
Step 1 handles the data of tested point:
It is similar with step 2 with off-line phase step 1 according to data processing module, the data of tested point are handled,
Obtain CSI amplitude information and RSS information.
Step 2 calculates the coding of CSI at tested point:
It is input with the CSI amplitude information of tested point, according to deep layer autocoder model trained in off-line phase
Middle encoder section obtains its corresponding CSI encoded information.
Step 3 generates the joint RSS and CSI fingerprint of tested point:
According to the CSI encoded information and RSS information of tested point, and can be added corresponding according to actual location scene
AP selection algorithm, the joint fingerprint vector for generating tested point according to the step of off-line phase four.
Step 4 obtains the prediction result of tested point:
The fingerprint vector of the joint RSS and CSI generated using in step 3 is as the location prediction based on deep neural network
The input of model, output are prediction coordinate.
Below with reference to experiment to it is of the invention be exactly that effect is explained in detail.
The experimental result display present invention positioning accuracy with higher in actual indoor environment of Fig. 3.Experiment scene
It is a 16m (length) × 7.76m (width) meeting room of Xian Electronics Science and Technology University's laboratory building, wherein reference point totally 27, phase
Adjacent spacing be 1.6m, test point totally 18, adjacent spacing 1.6m.When collecting data, 1200 are collected at each reference point
Sample collects 900 samples at each test point.
In order to verify the validity that the present invention promotes positioning accuracy, the indoor positioning technologies of Fig. 3 comparison can be divided into: be based on
The indoor orientation method of RSS fingerprint, the indoor orientation method based on CSI fingerprint and the indoor positioning side based on RSS Yu CSI fingerprint
Method.Specifically, KNN is a kind of indoor orientation method based on RSS fingerprint, judged according to the Euclidean distance of RSS fingerprint vector
The distance of two positions.FIFS and DeepFi is the indoor orientation method based on CSI fingerprint.FIFS is flat with the weighting of CSI amplitude
Mean value is location fingerprint, realizes positioning based on probabilistic method.In order to cope with the CSI data bulk redundancy information between different sub-carrier
The problem of bringing with noise for location algorithm, DeepFi is using greedy algorithm by the CSI information at each reference point to layer-by-layer
Training stack is limited Boltzmann machine, and the stage is limited Boltzmann machine calculating reconstruct by the stack of each reference point and misses on line
Difference, and probability of test point Yu the reference point is calculated with this, finally complete the positioning to test point.RandomForest+LDA with
Localization method of the SVM+LDA based on RSS Yu CSI fingerprint first pre-processes CSI using the dimension-reduction algorithm of this classics of LDA, so
Afterwards using after dimensionality reduction CSI and RSS information establish fingerprint base, finally by machine learning model such as SVM and RandomForest into
Row location prediction.
The fingerprint indoor orientation method of joint RSS and CSI proposed by the present invention based on deep learning is referred to as DADN.With
KNN, DeepFi, FIFS, SVM+LDA, RandomForest+LDA comparison, CDF curve, average localization error, median error are fixed
The Comparison of experiment results of the comparison of position error to standard deviation shows that DADN algorithm is able to ascend positioning accuracy and stability: on the one hand,
DADN algorithm joint CSI and RSS establishes fingerprint base, enriches position feature, has apparent gain to positioning accuracy.On the other hand,
DADN algorithm based on deep learning is for mapping complicated between study these radio signal characteristics of CSI and RSS and position coordinates
Ability it is stronger.
1 median error of table, mean error, the comparison of position error standard deviation
Location algorithm | Median error (m) | Mean error (m) | Position error standard deviation |
DADN | 0.5 | 0.717 | 0.356 |
RandomForest+LDA | 1.7 | 1.873 | 0.835 |
SVM+LDA | 1.8 | 1.966 | 0.688 |
FIFS | 2.5 | 2.668 | 1.511 |
DeepFi | 2.0 | 2.532 | 1.424 |
KNN | 2.1 | 2.38 | 1.539 |
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (7)
1. a kind of fingerprint indoor orientation method of the joint RSS and CSI based on deep learning, which is characterized in that described based on deep
The fingerprint indoor orientation method for spending the joint RSS and CSI of study includes off-line phase and tuning on-line stage;
Off-line phase collects and handles RSS and CSI information, and CSI of the training based on deep layer self-encoding encoder handles model, to reference
CSI coding at point, joint RSS and CSI coding building fingerprint base, divide sub fingerprint library according to AP selection algorithm;For each
Sub fingerprint library training one is based on deep neural network location prediction model;
The tuning on-line stage is according to the trained CSI processing model based on deep layer self-encoding encoder of off-line phase to tested point
CSI coding, calculates joint RSS and CSI fingerprint, based on deep neural network model prediction positioning.
2. the fingerprint indoor orientation method of the joint RSS and CSI based on deep learning, feature exist as described in claim 1
In, it is described based on deep learning joint RSS and CSI fingerprint indoor orientation method off-line phase the following steps are included:
Step 1, data collection and processing;
Step 2 designs the structure of deep layer autocoder model, including its depth according to the CSI amplitude information that processing obtains,
Neuron number and activation primitive, and the CSI based on deep layer self-encoding encoder handles model accordingly for training as unit of AP;
Step 3 encodes the CSI data at reference point according to trained encoder, by CSI coding and RSS information
Combination, establishes new fingerprint base;
Step 4, location prediction model of the training based on deep-neural-network, planned network structure and activation primitive, are missed with positioning
Difference is target, specially minimizes the mean square error of normalization the prediction coordinate and true normalized coordinate of output;With corresponding
The fingerprint vector in sub fingerprint library is input, and true normalized coordinate is output, and normalization physical location coordinate is output, training base
In the location prediction model of deep neural network.
3. the fingerprint indoor orientation method of the joint RSS and CSI based on deep learning, feature exist as claimed in claim 2
In data collection and the processing of the step 1 specifically include:
(1) indoors in positioning scene, coordinate system is established, reference point and test point is divided, collects number at each position respectively
According to and record its relative coordinate;It for the data file of each reference point, is handled, is used using MATLAB software
Read_bf_file function in CSITool reads data;
(2) it arranges and is obtained with processing data, the get_total_rss being utilized respectively in CSITool and get_scaled_csi function
It takes and separates RSS and CSI information.
4. the fingerprint indoor orientation method of the joint RSS and CSI based on deep learning, feature exist as described in claim 1
In the tuning on-line stage of the fingerprint indoor orientation method of the joint RSS and CSI based on deep learning includes following step
It is rapid:
Step 1 handles the data of tested point, according to data processing module, handles the data of tested point, obtains CSI width
Spend information and RSS information;
Step 2 calculates the coding of CSI at tested point, is input with the CSI amplitude information of tested point, instructs according in off-line phase
Encoder section in the deep layer autocoder model perfected obtains its corresponding CSI encoded information;
Step 3 generates the joint RSS and CSI location fingerprint vector of tested point, according to the CSI encoded information and RSS of tested point
Information, and according to actual location scene, corresponding AP selection algorithm is added, generates tested point according to the step of off-line phase four
Combination fingerprint vector;
Step 4 obtains the prediction result of tested point, using the fingerprint vector of joint RSS and the CSI coding of generation as based on deep
The input for spending the location prediction model of neural network exports to predict coordinate.
5. it is a kind of based on described in claim 1 based on deep learning joint RSS and CSI fingerprint indoor orientation method based on
The fingerprint indoor locating system of the joint RSS and CSI of deep learning, which is characterized in that the joint RSS based on deep learning
Fingerprint indoor locating system with CSI includes:
Data collection and processing module collect RSS and CSI information, handle model pair by the CSI based on deep layer autocoder
CSI data carry out dimensionality reduction and feature extraction is handled;
Fingerprint combination module, for handling the CSI coded combination of model by RSS and by the CSI based on deep layer autocoder,
Establish fingerprint base;
Location prediction module, the data for being collected according to tested point predict coordinate, to combine the fingerprint base of RSS Yu CSI information
Data set is input, and by the location prediction model based on deep neural network, obtained output is to predict coordinate.
6. fixed in a kind of fingerprint room using the joint RSS and CSI described in Claims 1 to 4 any one based on deep learning
The wireless communication system of position method.
7. fixed in a kind of fingerprint room using the joint RSS and CSI described in Claims 1 to 4 any one based on deep learning
The indoor locating system of position method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910518903.1A CN110381440A (en) | 2019-06-16 | 2019-06-16 | The fingerprint indoor orientation method of joint RSS and CSI based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910518903.1A CN110381440A (en) | 2019-06-16 | 2019-06-16 | The fingerprint indoor orientation method of joint RSS and CSI based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110381440A true CN110381440A (en) | 2019-10-25 |
Family
ID=68248843
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910518903.1A Pending CN110381440A (en) | 2019-06-16 | 2019-06-16 | The fingerprint indoor orientation method of joint RSS and CSI based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110381440A (en) |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110933625A (en) * | 2019-11-01 | 2020-03-27 | 武汉纺织大学 | Ultra-wideband fingerprint positioning method based on deep learning |
CN111209960A (en) * | 2020-01-06 | 2020-05-29 | 天津工业大学 | CSI system multipath classification method based on improved random forest algorithm |
CN111428819A (en) * | 2020-04-27 | 2020-07-17 | 北京工业大学 | CSI indoor positioning method based on stacked self-coding network and SVM |
CN111551893A (en) * | 2020-06-19 | 2020-08-18 | 天津大学 | Deep learning and neural network integrated indoor positioning method |
CN112147573A (en) * | 2020-09-14 | 2020-12-29 | 山东科技大学 | Passive positioning method based on amplitude and phase information of CSI (channel State information) |
CN112333642A (en) * | 2020-11-06 | 2021-02-05 | 南京邮电大学 | Indoor positioning method based on channel state information |
CN112653991A (en) * | 2020-12-23 | 2021-04-13 | 北京工业大学 | WLAN indoor positioning method of TebNet neural network model based on deep learning |
CN112712557A (en) * | 2020-12-17 | 2021-04-27 | 上海交通大学 | Super-resolution CIR indoor fingerprint positioning method based on convolutional neural network |
CN112924928A (en) * | 2021-01-26 | 2021-06-08 | 西安交通大学 | Indoor Wi-Fi multi-person detection method based on path separation |
CN113411741A (en) * | 2020-10-23 | 2021-09-17 | 湖南国天电子科技有限公司 | Partition fusion positioning method based on WiFi and geomagnetic fingerprints |
CN113453148A (en) * | 2021-06-25 | 2021-09-28 | 南通大学 | Indoor position fingerprint positioning method combining deep learning and weighted K-neighbor algorithm |
CN113518307A (en) * | 2021-04-22 | 2021-10-19 | 上海大学 | Position fingerprint positioning method based on CSI and crowdsourcing migration self-calibration updating |
CN113852908A (en) * | 2021-07-29 | 2021-12-28 | 上海大学 | Indoor positioning method based on antenna extension |
CN113923773A (en) * | 2021-11-11 | 2022-01-11 | 深圳大学 | Deep learning-based indoor positioning method and device |
CN114268919A (en) * | 2021-11-12 | 2022-04-01 | 北京航空航天大学 | Channel state information fingerprint positioning method based on weighted k nearest neighbor |
CN114710831A (en) * | 2022-03-10 | 2022-07-05 | 南京市地铁交通设施保护办公室 | RFID label positioning system based on deep learning |
CN114866119A (en) * | 2022-04-15 | 2022-08-05 | 电子科技大学长三角研究院(湖州) | Hybrid beam forming method under imperfect channel state information condition |
CN116233747A (en) * | 2023-03-09 | 2023-06-06 | 江苏亨鑫科技有限公司 | BLE positioning method and device based on transfer learning and fingerprint library updating |
CN116527462A (en) * | 2023-05-25 | 2023-08-01 | 兰州交通大学 | Wireless positioning method based on channel state change CSI value |
CN117062001A (en) * | 2023-08-01 | 2023-11-14 | 哈尔滨工程大学 | 5G NR indoor positioning method and system based on interpretable deep learning |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105163282A (en) * | 2015-09-22 | 2015-12-16 | 济南东朔微电子有限公司 | Indoor positioning system and positioning method based on Bluetooth location fingerprint |
CN105301558A (en) * | 2015-09-22 | 2016-02-03 | 济南东朔微电子有限公司 | Indoor positioning method based on bluetooth position fingerprints |
CN109633530A (en) * | 2018-11-30 | 2019-04-16 | 哈尔滨工业大学(深圳) | A kind of localization method and system, equipment, storage medium |
CN109640262A (en) * | 2018-11-30 | 2019-04-16 | 哈尔滨工业大学(深圳) | A kind of localization method and system, equipment, storage medium based on mixed-fingerprint |
-
2019
- 2019-06-16 CN CN201910518903.1A patent/CN110381440A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105163282A (en) * | 2015-09-22 | 2015-12-16 | 济南东朔微电子有限公司 | Indoor positioning system and positioning method based on Bluetooth location fingerprint |
CN105301558A (en) * | 2015-09-22 | 2016-02-03 | 济南东朔微电子有限公司 | Indoor positioning method based on bluetooth position fingerprints |
CN109633530A (en) * | 2018-11-30 | 2019-04-16 | 哈尔滨工业大学(深圳) | A kind of localization method and system, equipment, storage medium |
CN109640262A (en) * | 2018-11-30 | 2019-04-16 | 哈尔滨工业大学(深圳) | A kind of localization method and system, equipment, storage medium based on mixed-fingerprint |
Non-Patent Citations (3)
Title |
---|
CHAUR-HEH HSIEH ET.AL: "Deep Learning-Based Indoor Localization Using Received Signal Strength and Channel State Information", 《IEEE ACCESS》 * |
LINSHENG ZHAO ET.AL: "An Improved WiFi Indoor Localization Method Combining Channel State Information and Received Signal Strength", 《PROCEEDINGS OF THE 36TH CHINESE CONTROL CONFERENCE》 * |
唐伟华: "基于深度抽象的集成网络定位算法", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (29)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110933625A (en) * | 2019-11-01 | 2020-03-27 | 武汉纺织大学 | Ultra-wideband fingerprint positioning method based on deep learning |
CN111209960A (en) * | 2020-01-06 | 2020-05-29 | 天津工业大学 | CSI system multipath classification method based on improved random forest algorithm |
CN111209960B (en) * | 2020-01-06 | 2024-01-05 | 天津工业大学 | CSI system multipath classification method based on improved random forest algorithm |
CN111428819A (en) * | 2020-04-27 | 2020-07-17 | 北京工业大学 | CSI indoor positioning method based on stacked self-coding network and SVM |
CN111551893A (en) * | 2020-06-19 | 2020-08-18 | 天津大学 | Deep learning and neural network integrated indoor positioning method |
CN112147573A (en) * | 2020-09-14 | 2020-12-29 | 山东科技大学 | Passive positioning method based on amplitude and phase information of CSI (channel State information) |
CN113411741A (en) * | 2020-10-23 | 2021-09-17 | 湖南国天电子科技有限公司 | Partition fusion positioning method based on WiFi and geomagnetic fingerprints |
CN113411741B (en) * | 2020-10-23 | 2023-06-30 | 湖南国天电子科技有限公司 | Partition fusion positioning method based on WiFi and geomagnetic fingerprints |
CN112333642A (en) * | 2020-11-06 | 2021-02-05 | 南京邮电大学 | Indoor positioning method based on channel state information |
CN112712557A (en) * | 2020-12-17 | 2021-04-27 | 上海交通大学 | Super-resolution CIR indoor fingerprint positioning method based on convolutional neural network |
CN112653991A (en) * | 2020-12-23 | 2021-04-13 | 北京工业大学 | WLAN indoor positioning method of TebNet neural network model based on deep learning |
CN112924928B (en) * | 2021-01-26 | 2023-05-02 | 西安交通大学 | Indoor Wi-Fi multi-person detection method based on path separation |
CN112924928A (en) * | 2021-01-26 | 2021-06-08 | 西安交通大学 | Indoor Wi-Fi multi-person detection method based on path separation |
CN113518307A (en) * | 2021-04-22 | 2021-10-19 | 上海大学 | Position fingerprint positioning method based on CSI and crowdsourcing migration self-calibration updating |
CN113453148B (en) * | 2021-06-25 | 2022-05-13 | 南通大学 | Indoor position fingerprint positioning method combining deep learning and weighted K-neighbor algorithm |
CN113453148A (en) * | 2021-06-25 | 2021-09-28 | 南通大学 | Indoor position fingerprint positioning method combining deep learning and weighted K-neighbor algorithm |
CN113852908A (en) * | 2021-07-29 | 2021-12-28 | 上海大学 | Indoor positioning method based on antenna extension |
CN113852908B (en) * | 2021-07-29 | 2023-09-19 | 上海大学 | Indoor positioning method based on antenna extension |
CN113923773A (en) * | 2021-11-11 | 2022-01-11 | 深圳大学 | Deep learning-based indoor positioning method and device |
CN114268919B (en) * | 2021-11-12 | 2022-10-18 | 北京航空航天大学 | Channel state information fingerprint positioning method based on weighted k nearest neighbor |
CN114268919A (en) * | 2021-11-12 | 2022-04-01 | 北京航空航天大学 | Channel state information fingerprint positioning method based on weighted k nearest neighbor |
CN114710831A (en) * | 2022-03-10 | 2022-07-05 | 南京市地铁交通设施保护办公室 | RFID label positioning system based on deep learning |
CN114866119A (en) * | 2022-04-15 | 2022-08-05 | 电子科技大学长三角研究院(湖州) | Hybrid beam forming method under imperfect channel state information condition |
CN114866119B (en) * | 2022-04-15 | 2023-09-26 | 电子科技大学长三角研究院(湖州) | Mixed wave beam forming method under imperfect channel state information condition |
CN116233747A (en) * | 2023-03-09 | 2023-06-06 | 江苏亨鑫科技有限公司 | BLE positioning method and device based on transfer learning and fingerprint library updating |
CN116233747B (en) * | 2023-03-09 | 2023-12-22 | 江苏亨鑫科技有限公司 | BLE positioning method and device based on transfer learning and fingerprint library updating |
CN116527462A (en) * | 2023-05-25 | 2023-08-01 | 兰州交通大学 | Wireless positioning method based on channel state change CSI value |
CN116527462B (en) * | 2023-05-25 | 2024-02-02 | 兰州交通大学 | Wireless positioning method based on channel state change CSI value |
CN117062001A (en) * | 2023-08-01 | 2023-11-14 | 哈尔滨工程大学 | 5G NR indoor positioning method and system based on interpretable deep learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110381440A (en) | The fingerprint indoor orientation method of joint RSS and CSI based on deep learning | |
Hsieh et al. | Deep learning-based indoor localization using received signal strength and channel state information | |
Dai et al. | Combination of DNN and improved KNN for indoor location fingerprinting | |
CN103889051B (en) | Indoor WLAN fingerprint positioning methods based on AP ID filterings and Kalman filtering | |
Chai et al. | Reducing the calibration effort for location estimation using unlabeled samples | |
Adege et al. | Applying Deep Neural Network (DNN) for large-scale indoor localization using feed-forward neural network (FFNN) algorithm | |
CN109640262B (en) | Positioning method, system, equipment and storage medium based on mixed fingerprints | |
CN107727095A (en) | 3D indoor orientation methods based on spectral clustering and weighting reverse transmittance nerve network | |
Ghadimi et al. | Deep learning-based approach for low probability of intercept radar signal detection and classification | |
CN107743296A (en) | A kind of RSSI area segmentation formula localization methods based on compressed sensing | |
Ni et al. | An indoor pedestrian positioning method using HMM with a fuzzy pattern recognition algorithm in a WLAN fingerprint system | |
CN113747360B (en) | Indoor positioning method based on mixed visible light and BLE | |
Ezhumalai et al. | An efficient indoor positioning method based on Wi-Fi RSS fingerprint and classification algorithm | |
Cerar et al. | Improving CSI-based massive MIMO indoor positioning using convolutional neural network | |
CN113015093A (en) | Indoor wireless positioning method based on three-dimensional depth residual error neural network | |
CN113518307A (en) | Position fingerprint positioning method based on CSI and crowdsourcing migration self-calibration updating | |
Ma et al. | A novel indoor fingerprint localization system based on distance metric learning and AP selection | |
CN105866732A (en) | Improved MK model and WKNN algorithm combined mixed indoor positioning method | |
Liu et al. | CTSLoc: An indoor localization method based on CNN by using time-series RSSI | |
Chen et al. | A wifi indoor localization method based on dilated cnn and support vector regression | |
Yang et al. | Deep complex-valued convolutional neural network for drone recognition based on RF fingerprinting | |
CN112801268B (en) | Positioning method based on graph convolution and multilayer perceptron hybrid network | |
Tarekegn et al. | Applying long short-term memory (LSTM) mechanisms for fingerprinting outdoor positioning in hybrid networks | |
CN105704676A (en) | Method for improving fingerprint indoor positioning precision through employing signal time correlation | |
CN110830939B (en) | Positioning method based on improved CPN-WLAN fingerprint positioning database |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191025 |