Channel scene identification method based on genetic algorithm

Abstract

The invention relates to a channel scene identification method based on a genetic algorithm, and belongs to the technical field of communication. Firstly, channel impulse response is obtained through channel measurement, and channel characteristic parameters are extracted; secondly, coding the channel features to generate an initial population, and performing crossover and mutation operation on the channel features by adopting a genetic algorithm to generate a new channel feature subset; then, applying a maximum correlation and minimum redundancy criterion, and selecting a feature parameter with the maximum correlation with a channel scene from the candidate feature subset so as to reduce the redundancy of the feature parameter; and then, inputting the screened characteristic parameters into a back propagation neural network model for training and verification. And a back propagation neural network model learns the relationship between the feature parameters and the channel scene labels to realize automatic identification of LoS and NLoS scenes. According to the method, the problems of large feature parameter redundancy, small scene correlation and low recognition accuracy in channel scene recognition are solved.

Description

Technical Field

[0001] The invention belongs to the technical field of communications and relates to a channel scene recognition method based on a genetic algorithm. Background Art

[0002] The characteristics of wireless channels are affected by the physical propagation environment and scatterers in the scene. Different propagation environments will lead to different channel characteristics. Therefore, scene recognition of wireless channels is of great significance in channel modeling, real-time positioning and base station site selection. For example:

[0003] Different channel scenarios require different channel models. Accurately identifying channel scenarios helps to establish more accurate channel models and improve the performance of communication systems.

[0004] Positioning methods based on arrival angle or arrival time require line-of-sight propagation of wireless signals as a prerequisite, and channel scene recognition is the basis for accurate positioning.

[0005] The site selection of base stations needs to be differentiated according to open environments and densely obstructed environments to ensure coverage and save site construction costs.

[0006] Traditional channel scene recognition methods usually have the following limitations:

[0007] The recognition accuracy cannot be guaranteed and is easily affected by environmental changes.

[0008] A large number of channel parameters are required, which increases the computational complexity, and the redundant information will affect the accuracy and efficiency of recognition.

[0009] Existing methods have difficulty in achieving high-precision recognition in complex environments.

[0010] With the deepening of research on the application of machine learning in wireless channels, channel scene recognition methods based on machine learning have gradually emerged, such as support vector machines, back propagation neural networks, and random forests. These methods achieve automatic recognition by learning the relationship between channel features and scene labels.

[0011] Although the channel scene recognition method based on machine learning has made some progress, it still has the following shortcomings:

[0012] Existing channel scene recognition relies on channel features. The performance of the scene recognition algorithm is improved by adding channel features, but this will inevitably increase the computational complexity of the algorithm. At the same time, the introduction of redundant channel features will also reduce the accuracy of the scene recognition method.

[0013] Reasonable screening of channel features before building the algorithm can reduce the redundancy of channel features, increase the correlation between channel features and scenes, and make the scene recognition method have better generalization ability. Summary of the invention

[0014] In view of this, the purpose of the present invention is to provide a channel scene recognition method based on genetic algorithm. First, the channel impulse response is obtained through channel measurement and the channel characteristic parameters are extracted; secondly, the channel characteristics are encoded to generate an initial population, and the genetic algorithm is used to perform crossover and mutation operations on the channel characteristics to generate a new channel feature subset; then, the maximum correlation and minimum redundancy criterion is applied to select the feature parameters with the greatest correlation with the channel scene from the candidate feature subset to reduce the redundancy of the feature parameters; then, the selected feature parameters are input into the back propagation neural network model for training and verification. The back propagation neural network model realizes automatic recognition of LoS (line of sight) and NLoS (non-line of sight) scenes by learning the relationship between these feature parameters and channel scene labels.

[0015] In order to achieve the above object, the present invention provides the following technical solutions:

[0016] A channel scene recognition method based on genetic algorithm, the method comprising the following steps:

[0017] Step 1: Perform channel measurement on the channel scenario, obtain as many multipath signals as possible, and extract channel parameters including channel impulse response, maximum received power, received power kurtosis index, received power skewness index, power rise time, Ricean K factor, delay spread, arrival angle spread, intra-cluster delay spread, intra-cluster arrival angle spread, intra-cluster vertical arrival angle spread, cluster number k, and inter-cluster horizontal angle;

[0018] Step 2: Select initial channel characteristics according to the channel parameters obtained in step 1, encode the channel characteristics as 0 or 1, and form different channel characteristic subsets;

[0019] Step 3: Construct a fitness function and use the fitness function to evaluate the candidate channel feature subset;

[0020] Step 4: Execute the genetic algorithm to cross and mutate the channel feature sequence corresponding to each channel feature subset to obtain a new channel feature subset;

[0021] Step 5: Use the maximum correlation and minimum redundancy criteria to screen the channel feature subset;

[0022] Step 6: Use the best channel feature subset to train the back propagation neural network to obtain a channel scene recognition model;

[0023] Step 7: Input the channel features to be predicted into the channel scene recognition model and output the predicted channel type.

[0024] Further, the step 2 is specifically as follows:

[0025] Step 2-1: Define parameter set F, where F is the maximum received power max(|h| 2 ), received power kurtosis index κ, received power skewness index S, power rise time Δτ, Rice K factor K r , delay spread σ τ , Angle expansion Intra-cluster delay spread σ CDS , the horizontal arrival angle spread within the cluster σ CASA , vertical arrival angle spread within the cluster σ CESA , number of clusters k, horizontal angle within the cluster It consists of 12 channel features in total.

[0026] Step 2-2: Consider the channel features in the channel data set as genetic coding bits, perform binary coding on each coding bit, and use 1 and 0 to represent whether the feature is retained or eliminated;

[0027] Step 2-3: Generate M individuals {F1, F2, F3, ... F M}, the channel feature subsets identified in each LoS / NLoS scenario are different individuals in the population, F i It is composed of 12 strings of 0s and 1s, {F1, F2, F3, ... F M} is the initial group;

[0028] Further, the step three is specifically as follows:

[0029] Step 3-1: The i-th channel feature subset is F i , its fitness is defined as

[0030]

[0031] Where n is the number of channel features, y sim (F i (j)) and y(F i (j)) are F i The channel scene label and expected value of the j-th sample in; is the expected mean of the jth sample. The average fitness r ave Expressed as

[0032]

[0033] Solve the channel feature set {F1, F2, F3, ... F M} and retain the two channel feature subsets with the highest fitness.

[0034] Further, the step 4 is specifically as follows:

[0035] Step 4-1: Assume that the channel feature subset retained in step 3 is F i and F j , where the feature codes are F i =(a1,a2,a3,...a 12 ) and F j =(b1,b2,b3,...b 12 ). i and F j The intersection is performed at point c, which is expressed as

[0036]

[0037] c=ceil(α·L(F)+1) (4)

[0038] Among them, α is a random number in (0,1); L(F) is the length of F; c is the feature number for starting genetic operation; ceil(·) means rounding up. The two channel feature subsets are crossed at point c to obtain a new individual F i ′ and F j ′.

[0039] Step 4-2: Channel feature subset F at point d i To mutate, its value range is The process is then expressed as

[0040] F i ′={a1,a2,...,a′ d ,...,a 12} (5)

[0041]

[0042] in, and are the original features a d The maximum and minimum values ​​of ; λ is a random number in [0,1].

[0043] Further, the step five is specifically as follows:

[0044] Step 5-1: After the crossover mutation in step 4, the channel feature set {F1′, F2′, F3′, ..., F′ M}. Calculate the redundancy of each subset, taking the lth channel feature subset F l ′ as an example, the calculation is

[0045]

[0046] Among them, f i and y are the i-th channel characteristic parameters and their LoS or NLoS scene labels respectively; I(fi ,y) is the mutual information between the i-th channel feature parameter and the channel scene label y; |F l ′| is the number of selected features.

[0047] Step 5-2: Calculate the correlation of each subset, taking the lth channel feature subset F l ′ as an example, the calculation is

[0048]

[0049] Step 5-3: Combine formulas (7) and (8) to obtain the maximum correlation minimum redundancy criterion

[0050] maxΦ(D,R), Φ=DR (9)

[0051]

[0052] Where Φ is the mRMR value of the channel feature set, I(f1,y)} is the mutual information between the i-th channel feature parameter and the j-th channel feature parameter, and the largest feature Φ(f1,y) in Φ is selected. i ), add the selected feature set F l ″

[0053]

[0054] All channel features are sorted by correlation, and the five features with the greatest correlation are retained to form a new channel feature set.

[0055] Further, the step six is ​​specifically as follows:

[0056] Step 6-1: Encode the 12 channel features in the LoS and NLoS scenarios with 0 and 1 to generate the initial channel feature set {F1, F2, F3, ... F M};

[0057] Step 6-2: Add channel scene labels to the channel data samples, divide them into training set and test set evenly in proportion, and perform normalization;

[0058] Step 6-3: Initialize the parameters, set the population size of the genetic algorithm to 10, the maximum number of iterations to 20, the number of neurons in the input layer, hidden layer and output layer of the back propagation neural network to 12, 5 and 1 respectively, use the tansig activation function and Trainlm training algorithm, and the maximum number of iterations to 1000;

[0059] Step 6-4: Input the best channel feature subset to train the neural network model.

[0060] The beneficial effects of the present invention are that it can utilize the maximum relevance minimum redundancy criterion and genetic algorithm in channel scene recognition to efficiently screen out feature parameters that are highly relevant to the scene, significantly improving the accuracy and efficiency of recognition. The method uses a genetic algorithm to perform feature crossover and mutation operations, and selects the optimal feature parameters in combination with the maximum relevance minimum redundancy criterion, effectively reducing the redundancy of the feature parameters. By inputting the screened feature parameters into the back-propagation neural network model, automatic and accurate recognition of LoS and NLoS scenes is achieved. Experiments on 1745 sets of indoor scene channel simulation data sets show that compared with the genetic algorithm, the proposed algorithm improves the recognition accuracy by 8%, the efficiency by 38.8%, and shows superiority in convergence and migration generalization capabilities.

[0061] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0063] Figure 1 It is a flow chart of the channel scene recognition method based on genetic algorithm;

[0064] Figure 2 Schematic diagram of channel feature selection based on genetic algorithm;

[0065] Figure 3 Schematic diagram of the combination of back-propagation neural network training and channel feature selection.

[0066] Figure 4 is the fitness result of a single feature;

[0067] Figure 5 Simulation results for algorithm identification. DETAILED DESCRIPTION

[0068] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0069] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0070] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0071] like Figure 1 As shown, the channel scene recognition method based on genetic algorithm described in the present invention comprises the following steps:

[0072] Step 1: Perform channel measurement on the channel scenario, obtain as many multipath signals as possible, and extract channel parameters including channel impulse response, maximum received power, received power kurtosis index, received power skewness index, power rise time, Ricean K factor, delay spread, arrival angle spread, intra-cluster delay spread, intra-cluster arrival angle spread, intra-cluster vertical arrival angle spread, cluster number k, and inter-cluster horizontal angle;

[0073] Step 2: Select initial channel characteristics according to the channel parameters obtained in step 1, encode the channel characteristics as 0 or 1, and form different channel characteristic subsets;

[0074] Figure 2 Schematic diagram of channel feature selection based on genetic algorithm.

[0075] Step 3: Construct a fitness function and use the fitness function to evaluate the candidate channel feature subset;

[0076] Step 4: Execute the genetic algorithm to cross and mutate the channel feature sequence corresponding to each channel feature subset to obtain a new channel feature subset;

[0077] Step 5: Use the maximum correlation and minimum redundancy criteria to screen the channel feature subset;

[0078] Figure 3 Schematic diagram of the combination of back-propagation neural network training and channel feature selection.

[0079] Step 6: Use the best channel feature subset to train the back propagation neural network to obtain a channel scene recognition model;

[0080] Step 7: Input the channel features to be predicted into the channel scene recognition model and output the predicted channel type.

[0081] Further, the step 2 is specifically as follows:

[0082] Step 2-1: Define parameter set F, where F is the maximum received power max(|h| 2 ), received power kurtosis index κ, received power skewness index S, power rise time Δτ, Rice K factor K r , delay spread σ τ , Angle expansion Intra-cluster delay spread σ CDS , the horizontal arrival angle spread within the cluster σ CASA , vertical arrival angle spread within the cluster σ CESA , number of clusters k, horizontal angle within the cluster It consists of 12 channel features in total.

[0083] Step 2-2: Consider the channel features in the channel data set as genetic coding bits, perform binary coding on each coding bit, and use 1 and 0 to represent whether the feature is retained or eliminated;

[0084] Step 2-3: Generate M individuals {F1, F2, F3, ... F M}, the channel feature subsets identified in each LoS / NLoS scenario are different individuals in the population, F i It is composed of 12 strings of 0s and 1s, {F1, F2, F3, ... F M} is the initial group;

[0085] Further, the step three is specifically as follows:

[0086] Step 3-1: The i-th channel feature subset is F i , its fitness is defined as

[0087]

[0088] Where n is the number of channel features, y sim (F i (j)) and y(F i (j)) are F i The channel scene label and expected value of the j-th sample in; is the expected mean of the jth sample. The average fitness r ave Expressed as

[0089]

[0090] Solve the channel feature set {F1, F2, F3, ... F M} and retain the two channel feature subsets with the highest fitness.

[0091] Further, the step 4 is specifically as follows:

[0092] Step 4-1: Assume that the channel feature subset retained in step 3 is F i and F j , where the feature codes are F i =(a1,a2,a3,...a 12 ) and F j =(b1,b2,b3,...b 12 ). i and F j The intersection is performed at point c, which is expressed as

[0093]

[0094] c=ceil(α·L(F)+1) (4)

[0095] Among them, α is a random number in (0,1); L(F) is the length of F; c is the feature number for starting genetic operation; ceil(·) means rounding up. The two channel feature subsets are crossed at point c to obtain a new individual F i ′ and F j ′.

[0096] Step 4-2: Channel feature subset F at point d i To mutate, its value range is The process is then expressed as

[0097] F i ′={a1,a2,...,a′ d ,...,a 12} (5)

[0098]

[0099] in, and are the original features a d The maximum and minimum values ​​of ; λ is a random number in [0,1].

[0100] Further, the step five is specifically as follows:

[0101] Step 5-1: After the crossover mutation in step 4, the channel feature set {F1′, F2′, F3′, ..., F′ M}. Calculate the redundancy of each subset, taking the lth channel feature subset F l ′ as an example, the calculation is

[0102]

[0103] Among them, f i and y are the i-th channel characteristic parameters and their LoS or NLoS scene labels respectively; I(f i ,y) is the mutual information between the i-th channel feature parameter and the channel scene label y; |F l ′| is the number of selected features.

[0104] Step 5-2: Calculate the correlation of each subset, taking the lth channel feature subset F l ′ as an example, the calculation is

[0105]

[0106] Step 5-3: Combine formulas (7) and (8) to obtain the maximum correlation minimum redundancy criterion

[0107] maxΦ(D,R), Φ=DR (9)

[0108]

[0109] Where Φ is the mRMR value of the channel feature set, I(f1,y)} is the mutual information between the i-th channel feature parameter and the j-th channel feature parameter, and the largest feature Φ(f1,y) in Φ is selected. i ), add the selected feature set F l ″

[0110]

[0111] All channel features are sorted by correlation, and the five features with the greatest correlation are retained to form a new channel feature set.

[0112] Figure 4is the fitness result of a single feature. The channel features corresponding to the twelve feature indexes are: maximum received power, received power kurtosis index, received power skewness index, power rise time, Rice K factor, delay spread, arrival angle spread, intra-cluster delay spread, intra-cluster arrival angle spread, intra-cluster vertical arrival angle spread, cluster number k, and inter-cluster horizontal angle.

[0113] Figure 5 Simulation results for algorithm identification.

[0114] Further, the step six is ​​specifically as follows:

[0115] Step 6-1: Encode the 12 channel features in the LoS and NLoS scenarios with 0 and 1 to generate the initial channel feature set {F1, F2, F3, ... F M};

[0116] Step 6-2: Add channel scene labels to the channel data samples, divide them into training set and test set evenly in proportion, and perform normalization;

[0117] Step 6-3: Initialize the parameters, set the population size of the genetic algorithm to 10, the maximum number of iterations to 20, the number of neurons in the input layer, hidden layer and output layer of the back propagation neural network to 12, 5 and 1 respectively, use the tansig activation function and Trainlm training algorithm, and the maximum number of iterations to 1000;

[0118] Step 6-4: Input the best channel feature subset to train the neural network model.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A channel scene recognition method based on genetic algorithm, characterized by: The method comprises the following steps: Step 1: Perform channel measurement on the channel scenario, obtain several multipath signals, and extract channel parameters including channel impulse response, maximum received power, received power kurtosis index, received power skewness index, power rise time, Rice K factor, delay spread, arrival angle spread, intra-cluster delay spread, intra-cluster arrival angle spread, intra-cluster vertical arrival angle spread, cluster number k, and inter-cluster horizontal angle; Step 2: Select initial channel characteristics according to the channel parameters obtained in step 1, encode the channel characteristics as 0 or 1, and form different channel characteristic subsets; Step 3: Construct a fitness function and use the fitness function to evaluate the candidate channel feature subset; Step 4: Execute the genetic algorithm to cross and mutate the channel feature sequence corresponding to each channel feature subset to obtain a new channel feature subset; Step 5: Use the maximum correlation and minimum redundancy criteria to screen the channel feature subset; Step 6: Use the best channel feature subset to train the back propagation neural network to obtain a channel scene recognition model; Step 7: Input the channel features to be predicted into the channel scene recognition model and output the predicted channel type.

2. The channel scene recognition method based on genetic algorithm according to claim 1, characterized in that: The step 2 is specifically as follows: Step 2-1: Define parameter set F, where F is the maximum received power max(|h| 2 ), received power kurtosis index κ, received power skewness index S, power rise time Δτ, Rice K factor K r , delay spread σ τ , Angle expansion Intra-cluster delay spread σ CDS , the horizontal arrival angle spread within the cluster σ CASA , vertical arrival angle spread within the cluster σ CESA , number of clusters k, horizontal angle within the cluster It consists of 12 channel characteristic parameters in total; Step 2-2: Consider the channel features in the channel data set as genetic coding bits, perform binary coding on each coding bit, and use 1 and 0 to represent whether the feature is retained or eliminated; Step 2-3: Generate M individuals {F1, F2, F3, ... F M }, the channel feature subsets identified in each LoS / NLoS scenario are different individuals in the population, F i It is composed of 12 strings of 0s and 1s, {F1, F2, F3, ... F M } is the initial group.

3. The channel scene recognition method based on genetic algorithm according to claim 2, characterized in that: The step three is specifically as follows: Step 3-1: The i-th channel feature subset is F i , its fitness is defined as Where n is the number of channel features, y sim (F i (j)) and y(F i (j)) are F i The channel scene label and expected value of the j-th sample in; is the expected mean of the jth sample; the average fitness r ave Expressed as Solve the channel feature set {F1, F2, F3, ... F M } and retain the two channel feature subsets with the highest fitness.

4. The channel scene recognition method based on genetic algorithm according to claim 3, characterized in that: The step 4 is specifically as follows: Step 4-1: Assume that the channel feature subset retained in step 3 is F i and F j , where the feature codes are F i =(a1,a2,a3,...a 12 ) and F j =(b1,b2,b3,...b 12 ); i and F j The intersection is performed at point c, which is expressed as c=ceil(α·L(F)+1) (4) Where α is a random number in (0,1); L(F) is the length of F; c is the feature number at the start of the genetic operation; ceil(·) means rounding up; the two channel feature subsets are crossed at point c to obtain the new individual F i ′ and F j ′; Step 4-2: Channel feature subset F at point d i To mutate, its value range is The process is then expressed as <h2 style=";text-align:left;direction:ltr">F<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> ′={a1,a2,...,a′<h2 style=";text-align:left;direction:ltr"> d <h2 style=";text-align:left;direction:ltr"> ,...,a<h2 style=";text-align:left;direction:ltr"> 12 <h2 style=";text-align:left;direction:ltr">} (5) in, and are the original features a d The maximum and minimum values ​​of ; λ is a random number in [0,1].

5. The channel scene recognition method based on genetic algorithm according to claim 4, characterized in that: The step five is specifically as follows: Step 5-1: After the crossover mutation in step 4, the channel feature set {F1′, F2′, F3′, ..., F′ M }; Calculate the redundancy of each subset, taking the lth channel feature subset F l ′ as an example, the calculation is Among them, f i and y are the i-th channel characteristic parameters and their LoS or NLoS scene labels respectively; I(f i ,y) is the mutual information between the i-th channel feature parameter and the channel scene label y; |F l ′| is the number of selected features; Step 5-2: Calculate the correlation of each subset, taking the lth channel feature subset F l ′ as an example, the calculation is Step 5-3: Combine formulas (7) and (8) to obtain the maximum correlation minimum redundancy criterion maxΦ(D,R), Φ=DR (9) Where Φ is the mRMR value of the channel feature set, I(f1,y)} is the mutual information between the i-th channel feature parameter and the j-th channel feature parameter, and the largest feature Φ(f1,y) in Φ is selected. i ), add to the selected feature set F l ″ All channel features are sorted by correlation, and the five features with the greatest correlation are retained to form a new channel feature set.

6. The channel scene recognition method based on genetic algorithm according to claim 5, characterized in that: The step six is ​​specifically as follows: Step 6-1: Encode the 12 channel features in the LoS and NLoS scenarios with 0 and 1 to generate the initial channel feature set {F1, F2, F3, ... F M }; Step 6-2: Add channel scene labels to the channel data samples, divide them into training set and test set evenly in proportion, and perform normalization; Step 6-3: Initialize the parameters, set the population size of the genetic algorithm to 10, the maximum number of iterations to 20, the number of neurons in the input layer, hidden layer and output layer of the back propagation neural network to 12, 5 and 1 respectively, use the tansig activation function and Trainlm training algorithm, and the maximum number of iterations to 1000; Step 6-4: Input the best channel feature subset to train the neural network model.

Images