CN113260044A - CSI fingerprint positioning method, device and equipment based on double-layer dictionary learning - Google Patents

CSI fingerprint positioning method, device and equipment based on double-layer dictionary learning Download PDF

Info

Publication number
CN113260044A
CN113260044A CN202110390650.1A CN202110390650A CN113260044A CN 113260044 A CN113260044 A CN 113260044A CN 202110390650 A CN202110390650 A CN 202110390650A CN 113260044 A CN113260044 A CN 113260044A
Authority
CN
China
Prior art keywords
learning model
dictionary learning
csi
sample
dictionary
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.)
Granted
Application number
CN202110390650.1A
Other languages
Chinese (zh)
Other versions
CN113260044B (en
Inventor
刘雯
邓中亮
王旭
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing University of Posts and Telecommunications
Original Assignee
Beijing University of Posts and Telecommunications
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beijing University of Posts and Telecommunications filed Critical Beijing University of Posts and Telecommunications
Priority to CN202110390650.1A priority Critical patent/CN113260044B/en
Publication of CN113260044A publication Critical patent/CN113260044A/en
Application granted granted Critical
Publication of CN113260044B publication Critical patent/CN113260044B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W64/00—Locating users or terminals or network equipment for network management purposes, e.g. mobility management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W64/00—Locating users or terminals or network equipment for network management purposes, e.g. mobility management
    • H04W64/006—Locating users or terminals or network equipment for network management purposes, e.g. mobility management with additional information processing, e.g. for direction or speed determination

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Software Systems (AREA)
  • Medical Informatics (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Image Analysis (AREA)

Abstract

The embodiment of the invention provides a CSI fingerprint positioning method, a device and equipment based on double-layer dictionary learning, wherein when an instruction for positioning a target is obtained, channel state information of the target is collected and used as CSI data of the target channel state information; acquiring a first sparse code and a region label of the target CSI data based on a first dictionary learning model obtained through pre-training; acquiring a second sparse code of the first sparse code based on a second dictionary learning model obtained by pre-training and the region label, and acquiring a position label of the second sparse code as a position label of the target by using a classifier in the second dictionary learning model; the classifier is used for classifying according to the difference of the position labels of the sparse coding. The scheme can improve the positioning accuracy and reduce the storage pressure and the data processing complexity.

Description

CSI fingerprint positioning method, device and equipment based on double-layer dictionary learning
Technical Field
The invention relates to the technical field of fingerprint positioning, in particular to a CSI fingerprint positioning method, device and equipment based on double-layer dictionary learning.
Background
In wireless communication, the fingerprint positioning method has the characteristic of positioning without changing the hardware of the equipment, so the method is widely applied. In a specific application, a Channel State Information (CSI) fingerprint positioning method establishes a mapping relationship between a position label of each Reference Point (RP) in a certain area and signal characteristics indicated by Channel State Information of the Reference node according to the principle that an indoor environment is complex and different signal strength Information is formed at different positions by signal reflection and refraction, so that the position labels and the signal characteristics in the area are stored in a one-to-one correspondence manner to obtain a position fingerprint database. In this way, when a target, such as a certain terminal, is located, the channel state information of the target can be acquired, and the location tag matched with the channel state information of the target is determined from the location fingerprint database and is used as the location tag of the target in the area, so that the target can be located.
However, in order to map richer scene information, the channel state information tends to contain relatively many channel characteristics. Therefore, the channel state information is easily affected by noise, multipath, personnel movement and the like, and generates large fluctuation, so that the channel state information in the position fingerprint database and the channel state information acquired during positioning have obvious difference, the matching effect is affected, and the positioning accuracy is reduced. And, the channel state information dimension is higher, therefore, the position fingerprint database constructed by the traditional method has a huge scale, resulting in the problem of larger storage pressure and data processing complexity.
Disclosure of Invention
The embodiment of the invention aims to provide a CSI fingerprint positioning method, a device and equipment based on double-layer dictionary learning, so as to achieve the effects of improving positioning accuracy and reducing storage pressure and data processing complexity. The specific technical scheme is as follows:
in a first aspect, an embodiment of the present invention provides a CSI fingerprint positioning method based on two-layer dictionary learning, where the method includes:
when an instruction for positioning a target is acquired, acquiring channel state information of the target as CSI data of the target;
acquiring a first sparse code and a region label of the target CSI data based on a first dictionary learning model obtained through pre-training; the first dictionary learning model is used for distinguishing the region to which the CSI data belong based on a minimum reconstruction error principle, and is a sparse coding model obtained by training by utilizing a plurality of sample CSI data and a position label of each sample CSI data;
acquiring a second sparse code of the first sparse code based on a second dictionary learning model obtained by pre-training and the region label, and acquiring a position label of the second sparse code as a position label of the target by using a classifier in the second dictionary learning model;
the second dictionary learning model is used for enhancing the discrimination of the second sparse code based on a dictionary atom local constraint term and is a sparse code model obtained by utilizing the first sparse codes of the multiple sample CSI data and the position label training of each sample CSI data; the classifier is used for classifying according to the difference of the position labels of the sparse coding.
In a second aspect, an embodiment of the present invention provides a CSI fingerprint locating apparatus based on two-layer dictionary learning, where the apparatus includes:
the information acquisition module is used for acquiring channel state information of a target as CSI (channel state information) data of the target when an instruction for positioning the target is acquired;
the first coding module is used for acquiring a first sparse code and a region label of the target CSI data based on a first dictionary learning model obtained through pre-training; the first dictionary learning model is used for distinguishing the region to which the CSI data belong based on a minimum reconstruction error principle, and is a sparse coding model obtained by training by utilizing a plurality of sample CSI data and a position label of each sample CSI data;
the second coding module is used for acquiring a second sparse code of the first sparse code based on a second dictionary learning model obtained through pre-training and the region label, and acquiring a position label of the second sparse code by using a classifier in the second dictionary learning model to serve as the position label of the target;
the second dictionary learning model is used for enhancing the discrimination of the second sparse code based on a dictionary atom local constraint term and is a sparse code model obtained by utilizing the first sparse codes of the multiple sample CSI data and the position label training of each sample CSI data; the classifier is used for classifying according to the difference of the position labels of the sparse coding.
In a third aspect, an embodiment of the present invention provides an electronic device, where the electronic device includes a processor, a communication interface, a memory, and a communication bus, where the processor, the communication interface, and the memory complete communication with each other through the communication bus; a memory for storing a computer program; and the processor is used for realizing the steps of the CSI fingerprint positioning method based on the double-layer dictionary learning in the first aspect when the program stored in the memory is executed.
The embodiment of the invention has the following beneficial effects:
in the scheme provided by the embodiment of the invention, a first sparse code and a region label of target CSI data are obtained based on a first dictionary learning model obtained through pre-training, and a second sparse code of the first sparse code is obtained based on a second dictionary learning model and a region label obtained through pre-training. Therefore, the target CSI data can be sparsely encoded through the dictionary learning model, and concise expression and deep feature extraction of the high-latitude target CSI data are achieved, so that noise reduction is performed on the target CSI data, and storage pressure and data processing complexity are reduced. And, the model is learned through two layers of dictionaries: the first dictionary learning model and the second dictionary learning model respectively acquire second sparse codes of the region labels and the enhanced discrimination, and ensure that the sparse codes of different positioning regions have high discrimination, and the sparse codes of different fingerprint points, namely different position labels, have high discrimination, so that the positioning accuracy is improved. Based on the second sparse code, a classifier obtained through pre-training is utilized to obtain a position label of the second sparse code, and the position label is used as a position label of a target; the classifier is used for classifying according to the difference of the position labels of the sparse codes. Therefore, the matching process in fingerprint positioning can be simplified to classify the output of the second dictionary learning model, and the complexity of the matching process, namely data processing, can be reduced. Therefore, the scheme can improve the positioning accuracy and reduce the storage pressure and the data processing complexity.
Of course, not all of the advantages described above are necessarily required to practice any one product or method of the invention.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other embodiments can be obtained by using the drawings without creative efforts.
Fig. 1 is a schematic flowchart of a CSI fingerprint location method based on two-layer dictionary learning according to an embodiment of the present invention;
fig. 2 is an exemplary diagram of an application scenario of a CSI fingerprint location method based on two-layer dictionary learning according to an embodiment of the present invention;
fig. 3 is an exemplary diagram of a training process of a dictionary learning model in a CSI fingerprint positioning method based on two-layer dictionary learning according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of a CSI fingerprint locating apparatus based on two-layer dictionary learning according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived from the embodiments given herein by one of ordinary skill in the art, are within the scope of the invention.
Orthogonal Frequency Division Multiplexing (OFDM) is a bandwidth-limited digital multi-carrier modulation scheme in wireless communication, and has become the most widely used multi-carrier modulation technique. In Wi-Fi networks, OFDM, on the other hand, separates the signal into multiple orthogonal sub-channels with different frequencies, where the channel state information reflects characteristics of the communication link between the transmitter and receiver, including the effects of distance, scattering, fading, etc. on the signal. The channel state information may be expressed as:
Figure BDA0003016584010000041
wherein,
Figure BDA0003016584010000042
and
Figure BDA0003016584010000043
representing transmitted and received signal vectors, respectively, and the vectors
Figure BDA0003016584010000044
The H matrix represents the channel state information, which is the set of channel information of each subcarrier, and can be used
Figure BDA0003016584010000045
And
Figure BDA0003016584010000046
and (6) obtaining the estimation. Wherein H ═ H1,H2,...,Hn]TN is the number of subcarriers, Hi appears in complex form: hi＝|Hi|exp{j∠Hi}. Wherein, | HiI and HiRespectively amplitude and phase for the subcarrier i. Due to fading effects and frequency offset, the phase of the CSI data is compared to the amplitudeIt is slightly noisy, and therefore, noise can be reduced using the amplitude information of the CSI data.
Dictionary learning, also known as sparse representation, constructs a dictionary by learning a set of atoms and combining, such that a given signal can take good advantage of the sparse representation and can be reconstructed by a linear combination of several atoms in the dictionary matrix. Sparse coding refers to a process of obtaining a sparse signal x from an original signal y based on a dictionary D, and is generally expressed as the following constraint optimization objective:
Figure BDA0003016584010000047
where the equality constraint y Dx is too strict for the optimization problem, a smaller threshold may be used to relax the optimization problem. In addition, in order to make the algorithm more suitable for the classification problem, class label information or classifier parameters can be embedded into the objective function, so that the sparse coding has higher identification capability.
In combination with the above, an embodiment of the present invention provides a CSI fingerprint positioning method based on two-layer dictionary learning, and the method may be applied to an electronic device providing a positioning service. In a specific application, the electronic device may specifically include: desktop computers, portable computers, mobile terminals, wearable devices, internet televisions, and servers, among others.
As shown in fig. 1, a CSI fingerprint location method based on two-layer dictionary learning according to an embodiment of the present invention may include the following steps:
s101, when an instruction for positioning a target is acquired, acquiring channel state information of the target as CSI data of the target.
The manner of obtaining the instruction for positioning the target may be various. For example, an instruction for locating the target sent by the target may be received, or when the presence of the target is detected, it may be determined that the instruction for locating the target is obtained.
S102, acquiring a first sparse code and a region label of the target CSI data based on a first dictionary learning model obtained through pre-training.
The first dictionary learning model is used for distinguishing the region to which the CSI data belong based on a minimum reconstruction error principle, and is a sparse coding model obtained by training by utilizing a plurality of sample CSI data and a position label of each sample CSI data.
S103, acquiring a second sparse code of the first sparse code based on a second dictionary learning model and the region label obtained through pre-training, and acquiring a position label of the second sparse code as a position label of the target by using a classifier in the second dictionary learning model.
The second dictionary learning model is used for enhancing the discrimination of the second sparse code based on the dictionary atom local constraint term and is a sparse code model obtained by utilizing the first sparse codes of the multiple sample CSI data and the position label training of each sample CSI data; the classifier is used for classifying according to the difference of the position labels of the sparse coding.
In the scheme provided by the embodiment of the invention, the target CSI data can be sparsely encoded through the dictionary learning model, so that the concise expression and deep feature extraction of the high-latitude target CSI data are realized, the noise of the target CSI data is reduced, and the storage pressure and the data processing complexity are reduced. And, the model is learned through two layers of dictionaries: the first dictionary learning model and the second dictionary learning model respectively acquire second sparse codes of the region labels and the enhanced discrimination, and ensure that the sparse codes of different positioning regions have high discrimination, and the sparse codes of different fingerprint points, namely different position labels, have high discrimination, so that the positioning accuracy is improved. Moreover, the matching process in fingerprint positioning can be simplified into the step of classifying the output of the second dictionary learning model, so that the matching process, namely the complexity of data processing, can be reduced. Therefore, the scheme can improve the positioning accuracy and reduce the storage pressure and the data processing complexity.
In an optional implementation manner, the first dictionary learning model is obtained by training through the following steps:
acquiring sample data based on the plurality of sample CSI data and the position label of each sample CSI data;
inputting the sample data, the specification of the dictionary learning model and the initial parameters of the dictionary learning model into the target function of the first dictionary learning model, and performing iterative training on the target function of the first dictionary learning model:
when the target function of the first dictionary learning model is converged, taking the trained target function of the first dictionary learning model as the first dictionary learning model; the convergence comprises the following steps: the iteration times reach a first iteration threshold, or the difference of the output results between the target function of the current iteration and the target function of the last iteration is smaller than a first difference threshold;
when the target function of the first dictionary learning model is not converged, the updated first sparse code is obtained by using the sample data, and the target function of the trained first dictionary learning model is updated by using the updated first sparse code.
In an optional implementation manner, the second dictionary learning model is obtained by training through the following steps:
acquiring sample data based on the plurality of sample CSI data and the position label of each sample CSI data;
inputting a first sparse code of sample data, the specification of a dictionary learning model and initial parameters of the dictionary learning model into a target function of the first dictionary learning model, and performing iterative training on the target function of the first dictionary learning model:
when the target function of the second dictionary learning model is converged, taking the trained target function of the second dictionary learning model as the second dictionary learning model; the convergence comprises the following steps: the iteration times reach a second iteration threshold, or the difference of the output results between the target function of the current iteration and the target function of the last iteration is smaller than a second difference threshold;
when the target function of the second dictionary learning model is not converged, the updated second sparse code is obtained by using the plurality of sample CSI data and the graph Laplacian matrix of the position label of each sample CSI data, and the trained target function of the second dictionary learning model is updated by using the updated second sparse code.
Illustratively, as shown in fig. 2. The CSI fingerprint positioning method based on the double-layer dictionary learning provided by the embodiment of the invention can be regarded as comprising two stages: the first phase is an offline training phase and the second phase is an online positioning phase. The two alternative embodiments are the off-line training phase, and the embodiment of fig. 1 of the present invention is the on-line positioning phase. In the off-line training stage, the training data is Y, the dictionary specification is K, and the parameters are alpha, tau and lambda1,λ2And theta. This is done:
learning a model for a first dictionary: can pass through X(1)＝(D(11)TD(11)+τI+τΛ)-1D(11)TY updating sparse coding; by passing
Figure BDA0003016584010000071
The objective function of the trained first dictionary learning model is updated. Wherein, X(1)For the updated first sparse coding, D(11)Learning the objective function of the model for the trained first dictionary, D(12)Learning an objective function of the model for the updated trained first dictionary.
Learning the model for the second dictionary:
can pass through
Figure BDA0003016584010000072
Obtaining a graph Laplace matrix; through Z(21)＝(D(21)TD(21)+λ1L)-1D(21)Tx obtaining an updated second sparse code; by D(22)＝X(21)Z(21)T(Z(21)Z(21)T+Δ)-1Updating the dictionary in the second dictionary learning model, and by
Figure BDA0003016584010000073
And updating the classifier parameters.
Wherein L is a graph Laplacian matrix, D(21)Is trainedThe target function of the second dictionary learning model, delta is a diagonal matrix, the elements on the diagonal are Lagrange multipliers, D(22)Learning a dictionary in the model for the updated second dictionary, Z(21)For the updated second sparse coding, U ', b' are updated classifier parameters, n is the number of input sample CSI data, l is a loss function, UcIs a c-type hyperplane parameter, z, of a Support Vector Machine (SVM)iFor a second sparse coding of the ith sample CSI data,
Figure BDA0003016584010000081
for sample CSI data of the ith class of class c, bcIs the type c bias of the support vector machine.
In an optional implementation manner, the obtaining of the sample data based on the multiple sample CSI data and the location tag of each sample CSI data may specifically include the following steps:
respectively imaging each sample CSI data to obtain a plurality of CSI images;
aiming at each CSI image, acquiring the original average amplitude of each pixel point of the CSI image, and performing dispersion standardization on the original average amplitude of each pixel point to obtain a first standardized CSI image;
and obtaining sample data based on the CSI image after the first standardization.
For example, the manner of obtaining the sample data based on the first normalized CSI image may be various. In an alternative embodiment, the first normalized CSI image may be directly used as sample data. In another optional real-time manner, the obtaining of sample data based on the first normalized CSI image may specifically include the following steps:
and for each first normalized CSI image, performing dispersion normalization on corresponding pixel points in the first normalized CSI image by using the original average amplitude to obtain sample data.
Enhancing the positioning effect by using a CSI data track, and collecting CSI of each Reference Point (RP) for multiple times, for example, at the ith RP, grouping H CSI amplitude measurements from W subcarriers, thereby constructing an H × W matrix:
Figure BDA0003016584010000082
wherein,
Figure BDA0003016584010000083
is the CSI amplitude characteristic from the w-th subcarrier in the h-th timestamp of the ith RP. Since dictionary learning has excellent performance in coping with the task of image classification, CSI data can be visualized. First, a normalization process is applied to obtain clearer data from the CSI data matrix, i.e., each image is calculated at each location point liOriginal average amplitude of (d):
Figure BDA0003016584010000084
wherein A isiFor each image at each position point liThe original average amplitude of (d).
Dispersion normalization is used on each line of CSI images: for location point liRespectively recording the maximum value of each line
Figure BDA0003016584010000091
And minimum value
Figure BDA0003016584010000092
The normalization result of the single element of the h row and w column is
Figure BDA0003016584010000093
Figure BDA0003016584010000094
Finally, the normalized CSI amplitude image is
Figure BDA0003016584010000095
Figure BDA0003016584010000096
Also, to maintain the original output power of the image, each location point l may be referenced by the previously calculated average amplitudeiThe CSI image of (a) is normalized again:
Figure BDA0003016584010000097
wherein,
Figure BDA0003016584010000098
is a location point liThe normalized CSI image of (a) again,
Figure BDA0003016584010000099
is a location point liNormalized CSI image of (A)maxThe maximum amplitude of the CSI data in all location points.
In an optional implementation manner, the first dictionary learning model is:
Figure BDA00030165840100000910
wherein D is(1)For a first dictionary with sub-dictionaries aggregated, Z(1)A first sparse coding is performed on the target CSI data, n is the number of each region corresponding to the first dictionary, DlIs a sub-dictionary, X, belonging to the region l in the first dictionarylFor CSI data Y collected in region llIn the corresponding sub-dictionary DlThe above sparse coding, α, τ are constant parameters,
Figure BDA00030165840100000911
for dictionary atom constraint terms, f (D)l) Is composed ofA non-coherent promotion term for enhancing discrimination between CSI data belonging to different regions.
In an optional implementation manner, the second dictionary learning model is:
Figure BDA00030165840100000912
wherein Z is(2)For the second sparse coding of CSI data, U, b are both classifier parameters, λ1And λ2Are two constant parameters that are used to control,
Figure BDA0003016584010000101
locally constrained terms for dictionary atoms, D(2)In the form of a second dictionary of words,
Figure BDA0003016584010000102
for support of vector discrimination terms for discriminating between sparse codes belonging to different position tags, ucIs the normal vector associated with the class c hyperplane of the support vector represented by the support vector discriminant, bcIs the deviation corresponding to the support vector.
Illustratively, as shown in FIG. 3. The objective function in the training process is similar to the model, except that the parameters of the objective function are continuously updated in the training process until the parameters of the objective function are the same as the parameters of the model when the objective function converges. Therefore, the first dictionary learning model and the second dictionary learning model are both objective functions of converged dictionary learning models. In order to reduce the burden on data processing caused by a position fingerprint database containing a large number of fingerprint points, a region-specific sub-dictionary is trained in a first dictionary learning model. In order to have high discrimination between sparse coding of CSI amplitudes of different partitions, the following incoherent promoting terms are first introduced:
Figure BDA0003016584010000103
wherein D islSub-dictionary, X, corresponding to region llTraining for CSI acquired over region lData YlIn the corresponding sub-dictionary DlThe above sparse coding.
Figure BDA0003016584010000104
Represents XlI.e. excluding X from all training datalItself. Wherein, XlCorresponding to the partition l, i.e. training data YlCan be well covered by XlRepresents other than Xj(l ≠ j). As shown in the following formula,
Figure BDA0003016584010000105
in the optimization process, the size is as small as possible so as to ensure DlXjAnd YlNot close to this point, and this point will significantly increase the discrimination effect when the fingerprint points are partitioned:
Figure BDA0003016584010000106
and, in order to ensure XlAs sparse as possible, to reduce the existing model, most of the sparse coding will adopt l0Norm regularization or l1The computation time caused by norm regularization is used for training an efficient region classification model, and the embodiment of the invention utilizes l2，1Norm ensures row sparseness of coding coefficients, and l2，1The calculation of the norm is relatively easy. Finally, an objective function defining the first dictionary learning model is:
Figure BDA0003016584010000107
wherein,
Figure BDA0003016584010000108
and (4) keeping the calculation process of the first dictionary learning model stable for the dictionary atom constraint terms. The first dictionary learning model carries out region discrimination on the input CSI data without involving more detailed fingerprint matching, so that a classifier is not trained additionally, and the reconstruction error is the mostThe area discrimination is performed by using the miniaturization as a measurement standard:
Figure BDA0003016584010000109
that is, ynewPartitioned according to its sparse coding by assigning it to the class of objects that results in minimized reconstruction errors. In summary, in the two-layer learning model, the first dictionary learning model can ensure the sparse representation capability of the CSI data on the corresponding region sub-dictionary by introducing an incoherent promoting term.
In addition, in order to enable the sparse coding to have the fingerprint positioning capability, the first sparse coding of the first dictionary learning model can be used as the input of the second dictionary learning model, and a support vector discriminant is introduced, so that the sparse coding from different fingerprint points can be forced to be separated, which is specifically defined as:
Figure BDA0003016584010000111
and in addition, in consideration of sensitivity and instability of the CSI data, a dictionary atom local constraint term is introduced, so that the authenticity of the second dictionary learning model on the description of the CSI data manifold structure can be further enhanced. Finally, the objective function for the second layer is proposed as follows:
Figure BDA0003016584010000112
in this way, the sparse coding by the first dictionary learning model is input to the second dictionary learning, and the prediction of the position label is performed by the classifier parameters U and b.
Corresponding to the method embodiment, the embodiment of the invention also provides a CSI fingerprint positioning device based on double-layer dictionary learning.
As shown in fig. 4, an embodiment of the present invention provides a structure of a CSI fingerprint location apparatus based on two-layer dictionary learning, where the apparatus includes:
an information acquisition module 401, configured to acquire channel state information of a target as target channel state information CSI data when an instruction for positioning the target is acquired;
a first encoding module 402, configured to obtain a first sparse code and a region label of the target CSI data based on a first dictionary learning model obtained through pre-training; the first dictionary learning model is used for distinguishing the region to which the CSI data belong based on a minimum reconstruction error principle, and is a sparse coding model obtained by training by utilizing a plurality of sample CSI data and a position label of each sample CSI data;
a second encoding module 403, configured to obtain a second sparse code of the first sparse code based on a second dictionary learning model obtained through pre-training and the region tag, and obtain a position tag of the second sparse code as a position tag of the target by using a classifier in the second dictionary learning model;
the second dictionary learning model is used for enhancing the discrimination of the second sparse code based on a dictionary atom local constraint term and is a sparse code model obtained by utilizing the first sparse codes of the multiple sample CSI data and the position label training of each sample CSI data; the classifier is used for classifying according to the difference of the position labels of the sparse coding.
In the scheme provided by the embodiment of the invention, the target CSI data can be sparsely encoded through the dictionary learning model, so that the concise expression and deep feature extraction of the high-latitude target CSI data are realized, the noise of the target CSI data is reduced, and the storage pressure and the data processing complexity are reduced. And, the model is learned through two layers of dictionaries: the first dictionary learning model and the second dictionary learning model respectively acquire second sparse codes of the region labels and the enhanced discrimination, and ensure that the sparse codes of different positioning regions have high discrimination, and the sparse codes of different fingerprint points, namely different position labels, have high discrimination, so that the positioning accuracy is improved. Moreover, the matching process in fingerprint positioning can be simplified into the step of classifying the output of the second dictionary learning model, so that the matching process, namely the complexity of data processing, can be reduced. Therefore, the scheme can improve the positioning accuracy and reduce the storage pressure and the data processing complexity.
Optionally, the first dictionary learning model is obtained by training through the following steps:
acquiring sample data based on the plurality of sample CSI data and the position label of each sample CSI data;
inputting the sample data, the specification of the dictionary learning model and the initial parameters of the dictionary learning model into a target function of a first dictionary learning model, and performing iterative training on the target function of the first dictionary learning model:
when the target function of the first dictionary learning model is converged, taking the trained target function of the first dictionary learning model as the first dictionary learning model; the convergence comprises: the iteration times reach a first iteration threshold, or the difference of the output results between the target function of the current iteration and the target function of the last iteration is smaller than a first difference threshold;
and when the target function of the first dictionary learning model is not converged, acquiring updated first sparse codes by using the sample data, and updating the target function of the trained first dictionary learning model by using the updated first sparse codes.
An embodiment of the present invention further provides an electronic device, as shown in fig. 5, which includes a processor 501, a communication interface 502, a memory 503 and a communication bus 504, where the processor 501, the communication interface 502 and the memory 503 complete mutual communication through the communication bus 504,
a memory 503 for storing a computer program;
the processor 501, when executing the program stored in the memory 503, implements the following steps:
when an instruction for positioning a target is acquired, acquiring channel state information of the target as CSI data of the target;
acquiring a first sparse code and a region label of the target CSI data based on a first dictionary learning model obtained through pre-training; the first dictionary learning model is used for distinguishing the region to which the CSI data belong based on a minimum reconstruction error principle, and is a sparse coding model obtained by training by utilizing a plurality of sample CSI data and a position label of each sample CSI data;
acquiring a second sparse code of the first sparse code based on a second dictionary learning model obtained by pre-training and the region label, and acquiring a position label of the second sparse code as a position label of the target by using a classifier in the second dictionary learning model;
the second dictionary learning model is used for enhancing the discrimination of the second sparse code based on a dictionary atom local constraint term and is a sparse code model obtained by utilizing the first sparse codes of the multiple sample CSI data and the position label training of each sample CSI data; the classifier is used for classifying according to the difference of the position labels of the sparse coding.
In the scheme provided by the embodiment of the invention, the target CSI data can be sparsely encoded through the dictionary learning model, so that the concise expression and deep feature extraction of the high-latitude target CSI data are realized, the noise of the target CSI data is reduced, and the storage pressure and the data processing complexity are reduced. And, the model is learned through two layers of dictionaries: the first dictionary learning model and the second dictionary learning model respectively acquire second sparse codes of the region labels and the enhanced discrimination, and ensure that the sparse codes of different positioning regions have high discrimination, and the sparse codes of different fingerprint points, namely different position labels, have high discrimination, so that the positioning accuracy is improved. Moreover, the matching process in fingerprint positioning can be simplified into the step of classifying the output of the second dictionary learning model, so that the matching process, namely the complexity of data processing, can be reduced. Therefore, the scheme can improve the positioning accuracy and reduce the storage pressure and the data processing complexity.
The communication bus mentioned in the electronic device may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The communication bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is shown, but this does not mean that there is only one bus or one type of bus.
The communication interface is used for communication between the electronic equipment and other equipment.
The Memory may include a Random Access Memory (RAM) or a Non-Volatile Memory (NVM), such as at least one disk Memory. Optionally, the memory may also be at least one memory device located remotely from the processor.
The Processor may be a general-purpose Processor, including a Central Processing Unit (CPU), a Network Processor (NP), and the like; but also Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs) or other Programmable logic devices, discrete Gate or transistor logic devices, discrete hardware components.
In another embodiment of the present invention, a computer-readable storage medium is further provided, in which a computer program is stored, and the computer program, when executed by a processor, implements the steps of any of the above two-layer dictionary learning-based CSI fingerprint location methods.
In yet another embodiment of the present invention, there is also provided a computer program product containing instructions, which when run on a computer, causes the computer to perform any one of the above-mentioned two-layer dictionary learning-based CSI fingerprint location methods.
In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed on a computer, cause the processes or functions described in accordance with the embodiments of the invention to occur, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, from one website site, computer, server, or data center to another website site, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that incorporates one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk (SSD)), among others.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
All the embodiments in the present specification are described in a related manner, and the same and similar parts among the embodiments may be referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the apparatus and device embodiments, since they are substantially similar to the method embodiments, the description is relatively simple, and reference may be made to some descriptions of the method embodiments for relevant points.
The above description is only for the preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention shall fall within the protection scope of the present invention.

Claims (10)

1. A CSI fingerprint positioning method based on double-layer dictionary learning is characterized by comprising the following steps:
when an instruction for positioning a target is acquired, acquiring channel state information of the target as CSI data of the target;
acquiring a first sparse code and a region label of the target CSI data based on a first dictionary learning model obtained through pre-training; the first dictionary learning model is used for distinguishing the region to which the CSI data belong based on a minimum reconstruction error principle, and is a sparse coding model obtained by training by utilizing a plurality of sample CSI data and a position label of each sample CSI data;
acquiring a second sparse code of the first sparse code based on a second dictionary learning model obtained by pre-training and the region label, and acquiring a position label of the second sparse code as a position label of the target by using a classifier in the second dictionary learning model;
the second dictionary learning model is used for enhancing the discrimination of the second sparse code based on a dictionary atom local constraint term and is a sparse code model obtained by utilizing the first sparse codes of the multiple sample CSI data and the position label training of each sample CSI data; the classifier is used for classifying according to the difference of the position labels of the sparse coding.
2. The method of claim 1, wherein the first dictionary learning model is trained by the steps of:
acquiring sample data based on the plurality of sample CSI data and the position label of each sample CSI data;
inputting the sample data, the specification of the dictionary learning model and the initial parameters of the dictionary learning model into a target function of a first dictionary learning model, and performing iterative training on the target function of the first dictionary learning model:
when the target function of the first dictionary learning model is converged, taking the trained target function of the first dictionary learning model as the first dictionary learning model; the convergence comprises: the iteration times reach a first iteration threshold, or the difference of the output results between the target function of the current iteration and the target function of the last iteration is smaller than a first difference threshold;
and when the target function of the first dictionary learning model is not converged, acquiring updated first sparse codes by using the sample data, and updating the target function of the trained first dictionary learning model by using the updated first sparse codes.
3. The method of claim 1, wherein the second dictionary learning model is trained by the steps of:
acquiring sample data based on the plurality of sample CSI data and the position label of each sample CSI data;
inputting a first sparse code of the sample data, the specification of a dictionary learning model and initial parameters of the dictionary learning model into a target function of the first dictionary learning model, and performing iterative training on the target function of the first dictionary learning model:
when the target function of the second dictionary learning model is converged, taking the trained target function of the second dictionary learning model as the second dictionary learning model; the convergence comprises: the iteration times reach a second iteration threshold, or the difference of the output results between the target function of the current iteration and the target function of the last iteration is smaller than a second difference threshold;
when the objective function of the second dictionary learning model is not converged, obtaining updated second sparse codes by using the plurality of sample CSI data and the graph Laplacian matrix of the position label of each sample CSI data, and updating the objective function of the trained second dictionary learning model by using the updated second sparse codes.
4. The method according to any one of claims 2 to 3, wherein the obtaining sample data based on the plurality of sample CSI data and the location tag of each sample CSI data comprises:
respectively imaging each sample CSI data to obtain a plurality of CSI images;
aiming at each CSI image, acquiring the original average amplitude of each pixel point of the CSI image, and performing dispersion standardization on the original average amplitude of each pixel point to obtain a first standardized CSI image;
and obtaining the sample data based on the first normalized CSI image.
5. The method of claim 4, wherein obtaining the sample data based on the first normalized CSI image comprises:
and for each first normalized CSI image, performing dispersion normalization on corresponding pixel points in the first normalized CSI image by using the original average amplitude to obtain the sample data.
6. The method of claim 1, wherein the first dictionary learning model is:
Figure FDA0003016583000000031
wherein D is(1)For a first dictionary with sub-dictionaries aggregated, Z(1)A first sparse coding is performed on the target CSI data, n is the number of each region corresponding to the first dictionary, DlIs a sub-dictionary, X, belonging to the region l in the first dictionarylTo adopt in the region lSet of CSI data YlIn the corresponding sub-dictionary DlThe above sparse coding, α, τ are constant parameters,
Figure FDA0003016583000000032
for dictionary atom constraint terms, f (D)l) Is a non-coherent promoting term for enhancing discrimination between CSI data belonging to different regions.
7. The method of claim 1, wherein the second dictionary learning model is:
Figure FDA0003016583000000033
wherein Z is(2)For the second sparse coding of CSI data, U, b are both classifier parameters, λ1And λ2Are two constant parameters that are used to control,
Figure FDA0003016583000000034
locally constrained terms for dictionary atoms, D(2)In the form of a second dictionary of words,
Figure FDA0003016583000000035
for support of vector discrimination terms for discriminating between sparse codes belonging to different position tags, ucIs the normal vector associated with the c-th hyperplane of the support vector represented by the support vector discriminant, bcIs the deviation corresponding to the support vector.
8. An apparatus for locating CSI fingerprints based on two-layer dictionary learning, the apparatus comprising:
the information acquisition module is used for acquiring channel state information of a target as CSI (channel state information) data of the target when an instruction for positioning the target is acquired;
the first coding module is used for acquiring a first sparse code and a region label of the target CSI data based on a first dictionary learning model obtained through pre-training; the first dictionary learning model is used for distinguishing the region to which the CSI data belong based on a minimum reconstruction error principle, and is a sparse coding model obtained by training by utilizing a plurality of sample CSI data and a position label of each sample CSI data;
the second coding module is used for acquiring a second sparse code of the first sparse code based on a second dictionary learning model obtained through pre-training and the region label, and acquiring a position label of the second sparse code by using a classifier in the second dictionary learning model to serve as the position label of the target;
the second dictionary learning model is used for enhancing the discrimination of the second sparse code based on a dictionary atom local constraint term and is a sparse code model obtained by utilizing the first sparse codes of the multiple sample CSI data and the position label training of each sample CSI data; the classifier is used for classifying according to the difference of the position labels of the sparse coding.
9. The apparatus of claim 8, wherein the first dictionary learning model is trained by the following steps:
acquiring sample data based on the plurality of sample CSI data and the position label of each sample CSI data;
inputting the sample data, the specification of the dictionary learning model and the initial parameters of the dictionary learning model into a target function of a first dictionary learning model, and performing iterative training on the target function of the first dictionary learning model:
when the target function of the first dictionary learning model is converged, taking the trained target function of the first dictionary learning model as the first dictionary learning model; the convergence comprises: the iteration times reach a first iteration threshold, or the difference of the output results between the target function of the current iteration and the target function of the last iteration is smaller than a first difference threshold;
and when the target function of the first dictionary learning model is not converged, acquiring updated first sparse codes by using the sample data, and updating the target function of the trained first dictionary learning model by using the updated first sparse codes.
10. An electronic device is characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor and the communication interface are used for realizing mutual communication by the memory through the communication bus;
a memory for storing a computer program;
a processor for implementing the method steps of any of claims 1 to 7 when executing a program stored in the memory.
CN202110390650.1A 2021-04-12 2021-04-12 CSI fingerprint positioning method, device and equipment based on double-layer dictionary learning Active CN113260044B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110390650.1A CN113260044B (en) 2021-04-12 2021-04-12 CSI fingerprint positioning method, device and equipment based on double-layer dictionary learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110390650.1A CN113260044B (en) 2021-04-12 2021-04-12 CSI fingerprint positioning method, device and equipment based on double-layer dictionary learning

Publications (2)

Publication Number Publication Date
CN113260044A true CN113260044A (en) 2021-08-13
CN113260044B CN113260044B (en) 2022-04-05

Family

ID=77220634

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110390650.1A Active CN113260044B (en) 2021-04-12 2021-04-12 CSI fingerprint positioning method, device and equipment based on double-layer dictionary learning

Country Status (1)

Country Link
CN (1) CN113260044B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024050701A1 (en) * 2022-09-06 2024-03-14 华为技术有限公司 Positioning method, electronic device, computer-readable storage medium, and chip system

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170319098A1 (en) * 2016-05-06 2017-11-09 Case Western Reserve University System and method for spectroscopic magnetic resonance fingerprinting
CN109584270A (en) * 2018-11-13 2019-04-05 大连大学 Based on the visual tracking method for differentiating dictionary learning
CN110267194A (en) * 2019-06-06 2019-09-20 普联技术有限公司 A kind of radio roaming performance optimization method, device, storage medium and system
CN110889865A (en) * 2019-10-31 2020-03-17 广东技术师范大学 Video target tracking method based on local weighted sparse feature selection
CN111915830A (en) * 2020-07-24 2020-11-10 辽宁工程技术大学 Indoor intrusion detection method based on CSI signal sparse representation
CN112487231A (en) * 2020-12-17 2021-03-12 中国矿业大学(北京) Automatic image labeling method based on double-image regularization constraint and dictionary learning

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170319098A1 (en) * 2016-05-06 2017-11-09 Case Western Reserve University System and method for spectroscopic magnetic resonance fingerprinting
CN109584270A (en) * 2018-11-13 2019-04-05 大连大学 Based on the visual tracking method for differentiating dictionary learning
CN110267194A (en) * 2019-06-06 2019-09-20 普联技术有限公司 A kind of radio roaming performance optimization method, device, storage medium and system
CN110889865A (en) * 2019-10-31 2020-03-17 广东技术师范大学 Video target tracking method based on local weighted sparse feature selection
CN111915830A (en) * 2020-07-24 2020-11-10 辽宁工程技术大学 Indoor intrusion detection method based on CSI signal sparse representation
CN112487231A (en) * 2020-12-17 2021-03-12 中国矿业大学(北京) Automatic image labeling method based on double-image regularization constraint and dictionary learning

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
R. JHANSI RANI: "Latent Fingerprint Image Enhancement using Gabor Functions via Multi-Scale Patch based Sparse Representation and Matching based on Neural Networks", <2019 INTERNATIONAL CONFERENCE ON COMMUNICATION AND SIGNAL PROCESSING (ICCSP)> *
朱会杰等: "基于频域信号的稀疏编码在机械故障诊断中的应用", 《振动与冲击》 *
查培等: "基于字典学习的毫米波大规模MIMO系统混合预编码", 《电讯技术》 *
蒋天润尹露邓中亮: "信道状态信息指纹定位算法性能评价方法研究", 《导航定位与授时》 *
邓中亮等: "基于高精度室内位置感知的大数据研究与应用", 《计算机应用》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024050701A1 (en) * 2022-09-06 2024-03-14 华为技术有限公司 Positioning method, electronic device, computer-readable storage medium, and chip system

Also Published As

Publication number Publication date
CN113260044B (en) 2022-04-05

Similar Documents

Publication Publication Date Title
US10664381B2 (en) Method and system for synthetic generation of time series data
US11972213B2 (en) Event recognition method and apparatus, model training method and apparatus, and storage medium
US10609433B2 (en) Recommendation information pushing method, server, and storage medium
CN108280477B (en) Method and apparatus for clustering images
CN110472675B (en) Image classification method, image classification device, storage medium and electronic equipment
CN115953665B (en) Target detection method, device, equipment and storage medium
CN111612038B (en) Abnormal user detection method and device, storage medium and electronic equipment
CN108564102A (en) Image clustering evaluation of result method and apparatus
US20220417339A1 (en) Feature-based network embedding
CN114565807B (en) Method and device for training target image retrieval model
CN112131322B (en) Time sequence classification method and device
WO2023108995A1 (en) Vector similarity calculation method and apparatus, device and storage medium
CN113065525A (en) Age recognition model training method, face age recognition method and related device
CN111950647A (en) Classification model training method and device
CN113260044B (en) CSI fingerprint positioning method, device and equipment based on double-layer dictionary learning
US11941792B2 (en) Machine learning-based analysis of computing device images included in requests to service computing devices
CN114255381A (en) Training method of image recognition model, image recognition method, device and medium
US11755671B2 (en) Projecting queries into a content item embedding space
CN114842251A (en) Training method and device of image classification model, image processing method and device and computing equipment
US20220019986A1 (en) Vectorization of transactions
CN112085040B (en) Object tag determining method and device and computer equipment
CN113627514A (en) Data processing method and device of knowledge graph, electronic equipment and storage medium
CN113255824A (en) Method and device for training classification model and data classification
CN112861962A (en) Sample processing method, sample processing device, electronic device and storage medium
US20240104915A1 (en) Long duration structured video action segmentation

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