CN112564835A - 5G wireless channel multipath clustering calculation method based on KNN and SVM algorithm - Google Patents

5G wireless channel multipath clustering calculation method based on KNN and SVM algorithm Download PDF

Info

Publication number
CN112564835A
CN112564835A CN202011346164.1A CN202011346164A CN112564835A CN 112564835 A CN112564835 A CN 112564835A CN 202011346164 A CN202011346164 A CN 202011346164A CN 112564835 A CN112564835 A CN 112564835A
Authority
CN
China
Prior art keywords
clustering
algorithm
mpc
sample
mpcs
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
CN202011346164.1A
Other languages
Chinese (zh)
Other versions
CN112564835B (en
Inventor
温阳
赵雄文
杜飞
耿绥燕
周振宇
张磊
陈素红
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.)
State Grid Corp of China SGCC
Electric Power Research Institute of State Grid Shandong Electric Power Co Ltd
North China Electric Power University
Original Assignee
State Grid Corp of China SGCC
Electric Power Research Institute of State Grid Shandong Electric Power Co Ltd
North China Electric Power University
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 State Grid Corp of China SGCC, Electric Power Research Institute of State Grid Shandong Electric Power Co Ltd, North China Electric Power University filed Critical State Grid Corp of China SGCC
Priority to CN202011346164.1A priority Critical patent/CN112564835B/en
Publication of CN112564835A publication Critical patent/CN112564835A/en
Application granted granted Critical
Publication of CN112564835B publication Critical patent/CN112564835B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B17/00—Monitoring; Testing
    • H04B17/30—Monitoring; Testing of propagation channels
    • H04B17/391—Modelling the propagation channel
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B17/00—Monitoring; Testing
    • H04B17/30—Monitoring; Testing of propagation channels
    • H04B17/391—Modelling the propagation channel
    • H04B17/3912—Simulation models, e.g. distribution of spectral power density or received signal strength indicator [RSSI] for a given geographic region

Abstract

The invention provides a 5G wireless channel multipath clustering calculation method based on K-nearest neighbor (KNN) and Support Vector Machine (SVM) algorithm, which adopts a high-resolution channel parameter extraction algorithm to extract small-scale parameters in each snapshot of a channel; extracting a part of multipath components (MPC) by using a random sampling algorithm to perform clustering for pre-clustering; calculating a multidimensional relative distance of the extracted MPC by using a KNN algorithm and performing pre-clustering; carrying out pattern recognition on the pre-clustering labels of the known MPCs by using an SVM algorithm to obtain new clustering labels; and performing mode recognition on all MPCs by using the clustering labels generated by the SVM algorithm to obtain a final clustering result. The method can more accurately cluster the data of the wireless communication channel, thereby establishing a more accurate channel model and having very important application value for wireless channel link and system level performance simulation evaluation and network design under the 5G background.

Description

5G wireless channel multipath clustering calculation method based on KNN and SVM algorithm
Technical Field
The invention belongs to the technical field of wireless channel modeling, and particularly relates to a 5G wireless channel multipath clustering calculation method based on KNN and SVM algorithms.
Background
The precise channel model is the basis of the design and research of the wireless communication system, and in the field of wireless communication system research, the wireless channel model based on the cluster model is one of the most frequently used channel models. In 5G millimeter wave communication, the application of higher frequency band and larger antenna array enables higher resolution multipath Component (MPC) to be observed in the time delay domain and angle domain, so the clustering problem in 5G wireless channel research is more important. Clustering of MPCs helps to reduce the number of parameters in a Multiple-Input Multiple-Output (MIMO) channel model, and the cluster structure is more flexible in characterizing multilink scenarios.
There are several MPC clustering algorithms that consider the power, delay and angle information of multipath. However, due to the unsupervised learning attribute of the clustering algorithm, the initial value of the clustering algorithm must be determined according to the prior value, and the clustering needs to be further perfected by the supervised learning algorithm. The K-nearest neighbor (KNN) algorithm uses the K nearest neighbors to each sample to represent the sample value, thereby achieving sample classification. KNN is a relatively mature method in theory, but due to inertia of the KNN algorithm, the number of adjacent values of the KNN algorithm must be determined according to a priori value and manually adjusted according to different requirements. The Support Vector Machine (SVM) algorithm is a classifier with sparsity and robustness, and can implement multi-dimensional, high-density, non-linear clustering. When clustering is carried out, the SVM algorithm does not need to manually specify the number of data in a cluster, but needs a certain amount of labeled data for training.
Disclosure of Invention
Aiming at the technical problem, the invention provides a 5G wireless channel multipath component clustering calculation method based on KNN and SVM, which comprises the following steps:
step S101: extracting small-scale parameters in each snapshot of the channel by adopting a high-resolution channel parameter extraction algorithm;
step S102: extracting partial Multi-Path components (MPC) by using a random sampling algorithm to perform clustering to obtain a sample class for pre-clustering;
step S103: calculating a multidimensional relative distance for the MPC extracted in the step S102 by using a K-nearest neighbor (KNN) algorithm, performing pre-clustering processing, and respectively storing cluster labels obtained by pre-clustering in each MPC data;
step S104: performing mode recognition on the pre-clustering label of the MPC obtained in the step S103 by using a Support Vector Machine (SVM) algorithm to obtain a new clustering label;
step S105: and performing mode recognition on all MPCs by using the new clustering labels generated in the step S104 to obtain a final clustering result.
Preferably, the small-scale parameters include the number of MPCs, the MPC delay, the MPC horizontal arrival angle, and the MPC vertical arrival angle corresponding to each snapshot.
Preferably, the high resolution channel parameter extraction algorithm is a spatially alternating generalized expectation-maximization algorithm.
Further, the step S102 includes:
randomly rearranging all MPCs according to the small-scale parameters extracted in the step S101, and returning k random sample values as samples for pre-clustering;
calculating the geometric center point within the class:
Pcentral＝(xcentral，ycentral，zcentral，…)，
wherein x iscentral，ycentral，zcentral,., the average value of the samples of all dimensions is obtained, each multi-dimensional data is a different sample class, and only one multi-dimensional data exists in the class when the algorithm is started;
the Sum of the squared Errors (ESS) for each pre-clustered sample is calculated as follows:
Figure BDA0002799989100000021
enumerating two ESS values of all sample classes, and when the number of samples is k, the total number of ESS values is k (k-1)/2, and the calculation formula is as follows:
Figure BDA0002799989100000031
wherein, C1And C2Representing two sample classes, C1∪C2Is the union of the sample classes to be merged,
Figure BDA0002799989100000032
to calculate a new geometric center point from the binomial samples,
Figure BDA0002799989100000033
the distance from each point in the two-term sample to the new geometric center point;
compute merge C1And C2The total ESS value after two sample classes and the difference before merging:
Figure BDA0002799989100000034
and finding the two sample classes with the minimum cost increase of the combined ESS, combining the two sample classes to form a new sample class, and repeating the processes until k is reduced to 1 to obtain the sample class for pre-clustering.
Preferably, the value of the randomly sampled value k is half the number of all MPCs.
Further, in step 103, a normalized distance between any two extracted MPCs is calculated, and the distance result and the MPCs are used as input to train the KNN network, so as to obtain a pre-clustering label.
Further, the calculation formula of the euclidean distance between every two pre-clustered sample classes is as follows:
Figure BDA0002799989100000035
wherein n represents the highest dimension of the sample; x is the number ofkAnd ykA numerical value representing each dimension of the samples x and y;
and sorting the distances between each MPC and other MPCs according to an increasing relationship, selecting the K points with the smallest distance, determining the occurrence frequency of the category where the front K points are located, making a decision according to the dominant category in the occurrence frequencies, and returning the category with the highest occurrence frequency as a pre-clustering label of the MPC.
Further, the step S104 includes: substituting the extracted MPC and the pre-clustering label into an SVM algorithm to calculate a cost function J (·); calculating a regularization parameter gamma, and balancing a classification distance maximization and a relaxation variable by adjusting the value of the regularization parameter gamma to minimize a cost function; and calculating the position of the hyperplane in the multidimensional space through the pre-clustering label, and obtaining the clustering label through the partition of the hyperplane.
Further, an optimal classification hyperplane is determined to realize the training data obtained in step S103 as
Figure BDA0002799989100000036
Pattern recognition of (2), wherein xi∈RnIs the MPC small scale data, R, randomly extracted in step S102nRepresenting an n-dimensional space, yiIs the pre-clustering label obtained in step S103, k is the number of data samples;
the best classification hyperplane is defined as: h: omegaTφ(xi) + b is 0, where φ (x) is a kernel function and b is a constant which is fullFoot conditions:
yi[ωTφ(xi)+b]≥1-ei，ei＞0，i＝1，…，N
wherein e isiIs a relaxation variable, defined as ei> 0, i-1, …, N, the cost function at this point is calculated, i.e.:
Figure BDA0002799989100000041
where T denotes transposition, μ and ζ are superparameters for adjusting the sum of the regularization amount and the square error, and the parameter is adjusted with the ratio γ ═ ζ/μ serving as a regularization parameter;
the adjustment parameter of the cost function J (-) is minimized, and omega and e can be eliminated according to KKT condition (Karush-Kuhn-Tucker conditions)iThe following linear system is obtained as the optimal constraint:
Figure BDA0002799989100000042
wherein α ═ α1，…，αN]TIs alphaiTranspose of vector, I is an NxN dimensional identity matrix, and Ω is defined as Ωij＝φ(xi)Tφ(xj)＝K(xi，xj) Also known as kernel functions;
calculating a label discrimination function based on SVM pattern recognition as follows:
Figure BDA0002799989100000043
wherein the kernel function K (x)i，xj) Selecting Gauss radial basis function exp (- | | x)i-xj||2/2σ2)，
Figure BDA0002799989100000044
Clustering labels for the MPC small-scale parameters.
Further, the step S105 includes: training all MPCs according to the clustering labels obtained in the step S104, and determining an optimal classification surface and a clustering area in a multi-dimensional space; and (4) bringing all MPC small-scale data into a multi-dimensional space clustering region, determining a clustering label according to the region where the MPC small-scale data is located, and finishing clustering operation.
The invention has the beneficial effects that:
the 5G wireless channel multipath clustering algorithm based on the KNN and SVM algorithm can realize the clustering of multi-dimensional data, does not need accurate multipath prior information, can automatically adjust the clustering parameters through the superparameters, reduces the influence of the prior information on the clustering result, and improves the accuracy of distinguishing the clusters with higher MPC dimension and the clusters with similar spatial structures. Based on the method, the accurate clustering of the data of the 5G wireless communication channel can be realized, so that a more accurate channel model is established, and the method has very important application value for wireless channel link and system level performance simulation evaluation and network design under the 5G background.
Drawings
FIG. 1 is a block flow diagram of the method of the present invention;
FIG. 2 is a diagram illustrating the pre-clustering of step S103 according to the present invention;
FIG. 3 is a schematic diagram of the pattern recognition training of step S104 according to the present invention;
FIG. 4 is a schematic diagram of pattern recognition clustering in step S105 according to the present invention.
Detailed Description
The embodiments will be described in detail below with reference to the accompanying drawings.
As shown in fig. 1, the invention provides a 5G wireless channel multipath component clustering calculation method based on KNN and SVM, comprising:
step S101: extracting small-scale parameters in each snapshot of the channel by adopting a high-resolution channel parameter extraction algorithm:
and extracting small-scale parameters of the channel by using a Space-Alternative Generalized Expectation maximization (SAGE) algorithm according to actual measurement data of the channel. The SAGE algorithm is an expandable iterative algorithm of an Expectation Maximization (EM) algorithm, dimensionality is reduced by sequentially updating parameter subsets, the operation amount is reduced, and convergence speed is increased, so that parameter estimation is more accurate, and the signal-to-noise ratio of a system is improved. The small-scale parameters include the number of multi-paths, delay spread, horizontal direction angle, vertical direction angle, etc. corresponding to each snapshot.
Step S102: extracting part of MPCs by using a random sampling algorithm to perform clustering to obtain a sample class for pre-clustering;
and (4) randomly rearranging all the MPCs according to the small-scale parameters extracted in the step (S101), and returning k random sample values as samples for pre-clustering, wherein the value of k is preferably half of the number of all the MPCs.
Firstly, extracting pre-clustering samples from k random sample values, wherein the original data of each selected sample is multidimensional, each multidimensional data (preferably three-dimensional) is a different sample class, and the algorithm has only one multidimensional data in the class at the beginning. Calculating the geometric center point within the class:
Pcentral＝(xcentral，ycentral，zcentral，…)，
wherein x iscentral，ycentral，zcentral,.. is the sample average for each dimension.
The Sum of squared Errors (ESS) is defined as the geometric center point P from all the multi-dimensional sample classes to the class interiorcentralThe ESS of each pre-clustered sample is calculated as follows:
Figure BDA0002799989100000061
then enumerate two ESS values (when the number of samples is k, k x (k-1)/2 total ESS values) of all sample classes, which is calculated as follows:
Figure BDA0002799989100000062
wherein, C1And C2Representing two sample classes, C1∪C2Is the union of the sample classes to be merged,
Figure BDA0002799989100000063
to calculate a new geometric center point from the binomial samples,
Figure BDA0002799989100000064
the distance of each point in the binomial sample to the new geometric center point.
Compute merge C1And C2Total ESS value after sample class and difference before merging:
Figure BDA0002799989100000065
and finding the two sample classes with the minimum cost increase of the merged ESS, and merging the two sample classes to form a new sample class. The above process is repeated until k is reduced to 1, resulting in a sample class for pre-clustering.
Step S103: calculating the KNN distance for pre-clustering, wherein FIG. 2 is a schematic diagram of pre-clustering;
firstly, according to k multidimensional small-scale parameters randomly extracted in step S102, calculating Euclidean distance between every two pre-clustered sample classes, wherein the calculation formula is as follows:
Figure BDA0002799989100000066
wherein n represents the highest dimension of the sample; x is the number ofkAnd ykRepresenting the value of each dimension of the samples x and y.
And sorting the distances between each MPC and other MPCs according to an increasing relation, and selecting K points with the minimum distance. And determining the occurrence frequency of the category where the first K points are located, making a decision according to the dominant category in the occurrence frequencies, and returning the category with the highest occurrence frequency as a pre-clustering label of the MPC. In one embodiment of the invention, K has a value of 20.
Step S104: performing pattern recognition by using an SVM algorithm to obtain clustering labels, wherein FIG. 3 is a pattern recognition training schematic diagram;
the training data obtained according to step S103 is
Figure BDA0002799989100000071
Wherein xi∈RnIs the MPC small scale data, R, randomly extracted in step S102nRepresenting an n-dimensional space, yiIs the pre-cluster label found in step S103, and k is the number of data samples.
And determining the optimal classification hyperplane, namely realizing the pattern recognition of the training data. The best classification hyperplane is defined as: h: omegaTφ(xi) + b is 0, where Φ (x) is the kernel function and b is a constant, which satisfies the condition:
yi[ωTφ(xi)+b]≥1-ei，ei＞0，i＝1，…，N
wherein eiIs a relaxation variable, defined as ei> 0, i ═ 1, …, N. The original constraint condition is weakened after the relaxation variable is introduced, and the problem that the optimal classification hyperplane does not exist due to linear irreparability can be solved. The cost function at this time is calculated as:
Figure BDA0002799989100000072
where T denotes transposition, μ and ζ are superparameters for adjusting the sum of the regularization amount and the square error, and the parameter is adjusted with the ratio γ ═ ζ/μ serving as the regularization parameter.
The adjustment parameter of the cost function J (-) is minimized, and omega and e can be eliminated according to KKT condition (Karush-Kuhn-Tucker conditions)iThe following linear system is obtained as the optimal constraint:
Figure BDA0002799989100000073
wherein α ═ α1，…，αN]TIs alphaiTranspose of vector, I is an NxN dimensional identity matrix, and Ω is defined as Ωij＝φ(xi)Tφ(xj)＝K(xi，xj) Also known as kernel functions.
Calculating a label discrimination function based on SVM pattern recognition as follows:
Figure BDA0002799989100000081
wherein the kernel function K (x)i，xj) Selecting Gauss radial basis function exp (- | | x)i-xj||2/2σ2)，
Figure BDA0002799989100000082
Clustering labels for the MPC small-scale parameters.
Step S105: and performing mode recognition on all MPCs by using the clustering labels generated by the SVM algorithm to obtain a final clustering result, wherein FIG. 4 is a schematic diagram of mode recognition clustering.
And (5) training all the MPCs according to the clustering labels obtained in the step (S104) and determining an optimal classification surface and a clustering area in the multi-dimensional space. And (4) bringing all MPC small-scale data into a multi-dimensional space clustering region, determining a clustering label according to the region where the MPC small-scale data is located, and finishing clustering operation.
The present invention is not limited to the above embodiments, and any changes or substitutions that can be easily made by those skilled in the art within the technical scope of the present invention are also included in the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (10)

1. A5G wireless channel multipath clustering calculation method based on K neighbor and support vector machine algorithm is characterized by comprising the following steps:
step S101: extracting small-scale parameters in each snapshot of the channel by adopting a high-resolution channel parameter extraction algorithm;
step S102: extracting partial Multi-Path components (MPC) by using a random sampling algorithm to perform clustering to obtain a sample class for pre-clustering;
step S103: calculating a multidimensional relative distance for the MPC extracted in the step S102 by using a K-nearest neighbor (KNN) algorithm, performing pre-clustering processing, and respectively storing cluster labels obtained by pre-clustering in each MPC data;
step S104: performing mode recognition on the pre-clustering label of the MPC obtained in the step S103 by using a Support Vector Machine (SVM) algorithm to obtain a new clustering label;
step S105: and performing mode recognition on all MPCs by using the new clustering labels generated in the step S104 to obtain a final clustering result.
2. The method as claimed in claim 1, wherein the small-scale parameters include the number of MPCs, MPC delay, MPC horizontal arrival angle and MPC vertical arrival angle corresponding to each snapshot.
3. The method according to claim 1, wherein the high resolution channel parameter extraction algorithm is a space alternative generalized expectation-maximization algorithm.
4. The method for 5G wireless channel multipath clustering calculation based on K-nearest neighbor and support vector machine algorithm according to claim 1, wherein the step S102 comprises:
randomly rearranging all MPCs according to the small-scale parameters extracted in the step S101, and returning k random sample values as samples for pre-clustering;
calculating the geometric center point within the class:
Pcentral＝(xcentral,ycentral,zcentral,…)，
wherein x iscentral，ycentral，zcentral… is the sample average value of each dimension, each multidimensional data is a different sample class, and the algorithm has only one multidimensional data in the class at the beginning;
the Sum of the squared Errors (ESS) for each pre-clustered sample is calculated as follows:
Figure FDA0002799989090000021
enumerating two ESS values of all sample classes, and when the number of samples is k, the total number of ESS values is k (k-1)/2, and the calculation formula is as follows:
Figure FDA0002799989090000022
wherein, C1And C2Representing two sample classes, C1∪C2Is the union of the sample classes to be merged,
Figure FDA0002799989090000023
to calculate a new geometric center point from the binomial samples,
Figure FDA0002799989090000024
the distance from each point in the two-term sample to the new geometric center point;
compute merge C1And C2The total ESS value after two sample classes and the difference before merging:
Figure FDA0002799989090000025
and finding the two sample classes with the minimum cost increase of the combined ESS, combining the two sample classes to form a new sample class, and repeating the processes until k is reduced to 1 to obtain the sample class for pre-clustering.
5. The 5G wireless channel multipath clustering calculation method based on K-nearest neighbor and support vector machine algorithm as claimed in claim 4, wherein the value of the random sample value K is half of the number of all MPCs.
6. The method as claimed in claim 4, wherein in step 103, a normalized distance between any two extracted MPCs is calculated, and the distance result and the MPCs are used as input to train the KNN network to obtain the pre-clustering label.
7. The method for 5G wireless channel multipath clustering calculation based on K-nearest neighbor and support vector machine algorithm according to claim 6, characterized in that:
the calculation formula of Euclidean distance of every two pre-clustering sample classes is as follows:
Figure FDA0002799989090000031
wherein n represents the highest dimension of the sample; x is the number ofkAnd ykA numerical value representing each dimension of the samples x and y;
and sorting the distances between each MPC and other MPCs according to an increasing relationship, selecting the K points with the smallest distance, determining the occurrence frequency of the category where the front K points are located, making a decision according to the dominant category in the occurrence frequencies, and returning the category with the highest occurrence frequency as a pre-clustering label of the MPC.
8. The method for 5G wireless channel multipath clustering calculation based on K-nearest neighbor and support vector machine algorithm according to claim 6 or 7, wherein the step S104 comprises: substituting the extracted MPC and the pre-clustering label into an SVM algorithm to calculate a cost function J (·); calculating a regularization parameter gamma, and balancing a classification distance maximization and a relaxation variable by adjusting the value of the regularization parameter gamma to minimize a cost function; and calculating the position of the hyperplane in the multidimensional space through the pre-clustering label, and obtaining the clustering label through the partition of the hyperplane.
9. The method for 5G wireless channel multipath clustering calculation based on K-nearest neighbor and support vector machine algorithm according to claim 8, characterized in that:
determining the best classification hyperplane to realize the training data obtained in the step S103 as
Figure FDA0002799989090000032
Pattern recognition of (2), wherein xi∈RnIs the MPC small scale data, R, randomly extracted in step S102nRepresenting an n-dimensional space, yiIs the pre-clustering label obtained in step S103, k is the number of data samples;
the best classification hyperplane is defined as: h is omegaTφ(xi) + b is 0, where Φ (x) is the kernel function and b is a constant, which satisfies the condition:
yi[ωTφ(xi)+b]≥1-ei,ei＞0,i＝1,…,N
wherein e isiIs a relaxation variable, defined as ei> 0, i-1, …, N, the cost function at this point is calculated, i.e.:
Figure FDA0002799989090000041
where T denotes transposition, μ and ζ are superparameters for adjusting the sum of the regularization amount and the square error, and the parameter is adjusted with the ratio γ ═ ζ/μ serving as a regularization parameter;
the adjustment parameter of the cost function J (-) is minimized, and omega and e can be eliminated according to KKT condition (Karush-Kuhn-Tucker conditions)iThe following linear system is obtained as the optimal constraint:
Figure FDA0002799989090000042
wherein α ═ α1,…,αN]TIs alphaiTranspose of vector, I is an NxN dimensional identity matrix, and Ω is defined as Ωij＝φ(xi)Tφ(xj)＝K(xi,xj) Also known as kernel functions;
calculating a label discrimination function based on SVM pattern recognition as follows:
Figure FDA0002799989090000043
wherein the kernel function K (x)i,xj) Selecting Gauss radial basis function exp (- | | x)i-xj||2/2σ2)，
Figure FDA0002799989090000044
Clustering labels for the MPC small-scale parameters.
10. The method for 5G wireless channel multipath clustering calculation based on K-nearest neighbor and support vector machine algorithm according to claim 9, wherein the step S105 comprises: training all MPCs according to the clustering labels obtained in the step S104, and determining an optimal classification surface and a clustering area in a multi-dimensional space; and (4) bringing all MPC small-scale data into a multi-dimensional space clustering region, determining a clustering label according to the region where the MPC small-scale data is located, and finishing clustering operation.
CN202011346164.1A 2020-11-26 2020-11-26 5G wireless channel multipath clustering calculation method based on KNN and SVM algorithm Active CN112564835B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011346164.1A CN112564835B (en) 2020-11-26 2020-11-26 5G wireless channel multipath clustering calculation method based on KNN and SVM algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011346164.1A CN112564835B (en) 2020-11-26 2020-11-26 5G wireless channel multipath clustering calculation method based on KNN and SVM algorithm

Publications (2)

Publication Number Publication Date
CN112564835A true CN112564835A (en) 2021-03-26
CN112564835B CN112564835B (en) 2022-09-20

Family

ID=75045664

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011346164.1A Active CN112564835B (en) 2020-11-26 2020-11-26 5G wireless channel multipath clustering calculation method based on KNN and SVM algorithm

Country Status (1)

Country Link
CN (1) CN112564835B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115225180A (en) * 2022-07-18 2022-10-21 华北电力大学 5G wireless channel multipath clustering calculation method based on SVM-AKPD
WO2023020203A1 (en) * 2021-08-17 2023-02-23 中兴通讯股份有限公司 Wireless channel modeling method and apparatus, electronic device, and storage medium

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101295362A (en) * 2007-04-28 2008-10-29 中国科学院国家天文台 Combination supporting vector machine and pattern classification method of neighbor method
US20100217763A1 (en) * 2007-09-17 2010-08-26 Electronics And Telecommunications Research Institute Method for automatic clustering and method and apparatus for multipath clustering in wireless communication using the same
CN103400144A (en) * 2013-07-17 2013-11-20 山东师范大学 Active learning method based on K-neighbor for support vector machine (SVM)
CN107104747A (en) * 2017-06-20 2017-08-29 北京交通大学 The cluster-dividing method of multipath component in wireless time varying channel
US20180131575A1 (en) * 2016-11-07 2018-05-10 Beijing Jiaotong University Method for clustering wireless channel mpcs based on a kpd doctrine
CN108768567A (en) * 2018-06-01 2018-11-06 北京邮电大学 A kind of multipath cluster-dividing method, device, electronic equipment and readable storage medium storing program for executing
CN109299698A (en) * 2018-09-30 2019-02-01 北京交通大学 A kind of wireless channel scene recognition method based on support vector machines

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101295362A (en) * 2007-04-28 2008-10-29 中国科学院国家天文台 Combination supporting vector machine and pattern classification method of neighbor method
US20100217763A1 (en) * 2007-09-17 2010-08-26 Electronics And Telecommunications Research Institute Method for automatic clustering and method and apparatus for multipath clustering in wireless communication using the same
CN103400144A (en) * 2013-07-17 2013-11-20 山东师范大学 Active learning method based on K-neighbor for support vector machine (SVM)
US20180131575A1 (en) * 2016-11-07 2018-05-10 Beijing Jiaotong University Method for clustering wireless channel mpcs based on a kpd doctrine
CN107104747A (en) * 2017-06-20 2017-08-29 北京交通大学 The cluster-dividing method of multipath component in wireless time varying channel
CN108768567A (en) * 2018-06-01 2018-11-06 北京邮电大学 A kind of multipath cluster-dividing method, device, electronic equipment and readable storage medium storing program for executing
CN109299698A (en) * 2018-09-30 2019-02-01 北京交通大学 A kind of wireless channel scene recognition method based on support vector machines

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
QINGBI ZHENG: ""A Tracking-Based Multipath Components Clustering Algorithm"", 《2018 2ND URSI ATLANTIC RADIO SCIENCE MEETING(AT-RASC)》 *
YANG WEN: ""Machine Learning Based Mm-wave 60 GHz Channel Modeling for 5G Wireless Communication Systems"", 《2019 IEEE 5TH INTERNATIONAL CONFERENCE ON COMPUTER AND COMMUNICATIONS(ICCC)》 *
薛汉卿: ""面向5G信道的新型聚簇算法研究"", 《中国优秀博硕士学位论文全文数据库（硕士）信息科技辑》 *
赵雄文: ""基于最小二乘支持向量机的时变信道建模"", 《北京邮电大学学报》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023020203A1 (en) * 2021-08-17 2023-02-23 中兴通讯股份有限公司 Wireless channel modeling method and apparatus, electronic device, and storage medium
CN115225180A (en) * 2022-07-18 2022-10-21 华北电力大学 5G wireless channel multipath clustering calculation method based on SVM-AKPD

Also Published As

Publication number Publication date
CN112564835B (en) 2022-09-20

Similar Documents

Publication Publication Date Title
Balasko et al. Fuzzy clustering and data analysis toolbox
CN107241700B (en) Indoor positioning method based on CSI space-frequency characteristic and reference point position clustering algorithm
Brand et al. A unifying theorem for spectral embedding and clustering
CN110381436B (en) Rapid fingerprint positioning method based on large-scale MIMO single station system
CN112564835B (en) 5G wireless channel multipath clustering calculation method based on KNN and SVM algorithm
Al-Razgan et al. Weighted clustering ensembles
Lee et al. A novel clustering-based approach of indoor location fingerprinting
CN105472733B (en) It is a kind of applied in indoor positioning based on AP selective positioning method
CN110072186B (en) Weighted naive Bayes indoor positioning method based on independent attributes
CN109299698B (en) Wireless channel scene recognition method based on support vector machine
CN108171012B (en) Gene classification method and device
CN112468249A (en) 5G wireless channel multipath clustering algorithm based on adaptive nuclear power density
Wu et al. 3D CNN-enabled positioning in 3D massive MIMO-OFDM systems
Du et al. SVM-assisted adaptive kernel power density clustering algorithm for millimeter wave channels
Taner et al. Channel Charting in Real-World Coordinates
Kwedlo A new method for random initialization of the EM algorithm for multivariate Gaussian mixture learning
CN111523571B (en) Semi-supervised learning positioning method for distributed large-scale multi-antenna system
CN109871768B (en) Hyperspectral optimal waveband selection method based on shared nearest neighbor
CN114269014B (en) Large-scale MIMO dynamic environment fingerprint positioning method based on domain adaptive network
Muthukannan et al. Color image segmentation using k-means clustering and optimal fuzzy C-means clustering
US8943005B2 (en) Metric learning apparatus
Zhou et al. An optimized space partitioning technique to support two-layer WiFi fingerprinting
Boas et al. Deep-large: Higher-order svd and deep learning for model order selection in mimo ofdm systems
Taştan et al. Robust spectral clustering: A locality preserving feature mapping based on M-estimation
CN113473373B (en) UWB indoor positioning method

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