A wireless communication network relay selection method based on variational autoencoder

Abstract

The present invention discloses a wireless communication network relay selection method based on variational autoencoder, comprising the following steps: collecting a data set, establishing a variational autoencoder model, inputting the sampled data set into a VAE model to obtain an encoded data set and a decoded data set, and selecting an optimal relay node by importance sampling; the present invention can more accurately evaluate the performance of relay nodes and select the optimal relay node for signal forwarding, thereby improving the performance and reliability of the network.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless communication networks, and more particularly to a wireless communication network relay selection method based on variational autoencoder. Background Art

[0002] In wireless communication networks, relay node selection is a crucial task, which aims to improve the performance and reliability of the network by selecting appropriate relay nodes to forward signals. Existing relay node selection methods have some problems, such as inaccurate signal quality estimation and lack of flexibility in selecting relay nodes, so new solutions are needed.

[0003] In this field, there have been some studies that use machine learning techniques to solve the relay node selection problem. However, these methods often require a large amount of data for training, and may encounter problems such as insufficient training data and high computational complexity in practical applications. Therefore, a more efficient and accurate relay node selection method is needed.

[0004] In traditional wireless communication network relay selection methods, only the physical characteristics of transmission are often considered, such as signal strength, interference, and path loss. However, these methods ignore other factors that may affect relay selection, such as network topology, user demand, and device energy consumption. By introducing VAE, we can take these diverse factors into account and realize intelligent relay selection by learning data in the network.

[0005] Therefore, how to more accurately evaluate the performance of relay nodes and select the optimal relay node for signal forwarding, thereby improving the performance and reliability of the network, is a problem that urgently needs to be solved in this field. Summary of the invention

[0006] In view of this, the present invention provides a wireless communication network relay selection method based on variational autoencoder, which can more accurately evaluate the performance of relay nodes and select the optimal relay node for signal forwarding, thereby improving the performance and reliability of the network.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] A wireless communication network relay selection method based on variational autoencoder comprises the following steps:

[0009] Receive wireless communication network signals;

[0010] Preprocessing the wireless communication network signal to obtain initial data;

[0011] Inputting the initial data into a pre-built variational autoencoder model to obtain an encoded first data set;

[0012] Obtaining a set of candidate relay nodes by clustering the first data set;

[0013] Importance sampling is performed in the candidate relay node set to obtain the optimal relay node.

[0014] Furthermore, the importance sampling is specifically:

[0015] One or more evaluation indicators are confirmed for comprehensive evaluation; the evaluation indicators include bit error rate and / or signal-to-noise ratio.

[0016] Furthermore, the training steps of the variational autoencoder model are:

[0017] For the training data set X = {x 1 ,…,x N}, the goal is to maximize the marginal likelihood:

[0018]

[0019] Due to the non-optimizability of the marginal likelihood, when a model is parameterized using a neural network, an inference model (encoder) is usually introduced to optimize the lower bound of the variables:

[0020] lnp(x)≥E q(z|x) [lnp(x|z)]-KL(q(z|x)||p(z)) (2)

[0021] Among them, x represents the input data point, z represents the latent variable, q ( z|x ) represents the approximate posterior probability of the latent variable, E q(z|x) represents the expectation under the posterior distribution q(z|x), p(x|z) represents the decoder network, p(z)=Ν(z|0,I) represents the prior probability distribution; for continuous z, this is achieved by reparameterizing q(z|x).

[0022] Furthermore, by clustering the first data set, a set of candidate relay nodes is obtained, which is specifically:

[0023] Obtain all data points in the first data set;

[0024] The aggregation degree of each data point in the first data set is calculated, and it is iterated according to the aggregation degree to obtain multiple cluster centers; the data points corresponding to the cluster centers are the optional relay nodes.

[0025] Furthermore, the degree of aggregation is determined by a preference parameter, and the larger the preference parameter is, the stronger the degree of aggregation is;

[0026] The calculation formula of the preference parameter is:

[0027] p=median(s(k,i)),k,i=1,2,…,N

[0028] Where s(k,i) represents the data point x k With data point x i The similarity between the data points x k Preference parameters.

[0029] Furthermore, before iteration, the same preference parameters are set for each of the data points.

[0030] Furthermore, importance sampling is performed in the candidate relay node set, specifically: sampling the candidate relay node set according to a preset node distribution to generate a distribution sample set;

[0031] Calculating the mathematical expectation of each random variable in the distribution sample set;

[0032] The random variable with the largest mathematical expectation is extracted, and the sampling point corresponding to the random variable is used as the optimal relay node.

[0033] Furthermore, the preset node distribution is:

[0034]

[0035] Among them, p(x) is the preset node distribution, π(x) is the target distribution of the node, and f(x) is the distribution characteristic function of the distribution π(x).

[0036] Furthermore, it is characterized in that the calculation formula of the mathematical expectation is:

[0037]

[0038] in, is the importance sampling correction factor.

[0039] By sampling {x 1 ,x 2 ,…,x i ,…,x N}, and calculate the value of the function f(X) at each sample point {f(x 1 ),f(x 2 ),…,f(x i ),…,f(x N)}, and then calculate the sample mean of these function values ​​to estimate the mathematical expectation of the random variable f(X) under the p(x) distribution.

[0040] Beneficial effects of the present invention:

[0041] It can be known from the above technical solutions that, compared with the prior art, the present invention discloses a wireless communication network relay selection method based on variational autoencoder, which can more accurately evaluate the performance of relay nodes and select the optimal relay node for signal forwarding, thereby improving the performance and reliability of the network. The method has high practicality and market potential and can be widely used in various wireless communication network fields. By modeling the signal quality using the VAE model, the estimation accuracy of the signal quality can be improved, thereby more accurately evaluating the performance of the relay node. By analyzing the encoded data set, a set of candidate relay nodes can be obtained, thereby more accurately selecting the optimal relay node for signal forwarding. Using the importance sampling method, the performance of the optional relay nodes can be evaluated, thereby selecting the optimal relay node for signal forwarding. The optimal relay node can be quickly selected for signal forwarding in a real-time scenario, reducing the network's latency and energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0043] Figure 1 A schematic diagram of a wireless communication network relay selection method based on a variational autoencoder provided by the present invention;

[0044] Figure 2 This is a network structure diagram of the variational autoencoder in the present invention. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] The embodiment of the present invention discloses a wireless communication network relay selection method based on a variational autoencoder, comprising the following steps:

[0047] S1: Receive wireless communication network signals;

[0048] S2: Preprocess the wireless communication network signal to obtain initial data; the signal received by the relay node may be affected by many factors, such as noise, multipath effect, etc., so it is necessary to sample the received signal to obtain a representative data set; the sampling process adopts a random sampling method to ensure the adequacy and accuracy of the sampled data set;

[0049] S3: Input the initial data into a pre-built variational autoencoder model, such as Figure 2 The structure of the variational autoencoder model is shown. During the training process, the model will receive input data and map it to a latent representation in the latent space through the encoder. The mean μ and the variance σ are usually represented by two vectors. The latent variable z is sampled from the mean and variance vectors through the reparameterization technique, that is, z = μ + ε * σ, where ε is random noise sampled from a standard normal distribution. The encoded first data set is obtained; wherein, for the variational autoencoder model (VAE): VAE is a generative model that can encode input data into latent variables and decode them to generate new data; in the present invention, the VAE model is used to model the signal quality, convert the signal quality into latent variables, and generate new signal quality based on the latent variables.

[0050] By inputting the preprocessed first data set into the VAE model, an encoded data set and a decoded first data set are obtained; the encoded data set reflects the potential variables of signal quality and is used to analyze the set of candidate relay nodes; by performing a clustering analysis method on the encoded first data set, a set of candidate relay nodes can be obtained.

[0051] S4: By clustering the first data set, a set of candidate relay nodes is obtained; the clustering process is: obtaining all data points in the first data set; calculating the degree of aggregation of each data point in the first data set, and iterating according to the degree of aggregation to obtain multiple cluster centers; the data point corresponding to the cluster center is the optional relay node.

[0052] S5: Perform importance sampling in the candidate relay node set to obtain the optimal relay node. After obtaining the candidate relay node set according to the analysis results, importance sampling can be used to evaluate the performance of the optional relay nodes and select the optimal relay node for signal forwarding; when evaluating the performance of the optional relay nodes, various evaluation indicators such as bit error rate, signal-to-noise ratio, etc. can be used to evaluate the performance of the relay nodes; finally, the relay node with the best performance is selected for signal forwarding to improve the performance and reliability of the network.

[0053] In one embodiment, in S3, the training steps of the variational autoencoder model are:

[0054] For the training data set X = {x 1 ,…,x N}, that is, the set of nodes. The goal of the variational autoencoder is to maximize the marginal likelihood probability of the node distribution and obtain latent variables containing information reflecting the signal quality, which are used to analyze the candidate relay node set. The marginal likelihood probability solution is shown in formula (1):

[0055]

[0056] Due to the non-optimizability of marginal likelihood, when a neural network is used to parameterize the model, an encoder is usually introduced to optimize the lower limit of the variable. Formula (1) is further optimized as shown in Formula (2):

[0057]

[0058] Among them, x i represents the input data point, z represents the latent variable, q(z|x i ) represents the approximate posterior probability of the latent variable, Indicates that in the posterior distribution q(z|x i ) to find the expectation, p(x i |z) represents the decoder network, p(z) = Ν(z|0,I) represents the prior probability distribution; for continuous z, we reparameterize q(z|x i ) to achieve. Maximizing formula (2), to minimize q(z|x i ) and p(z), while maximizing the first term on the right side of the above equation.

[0059] In one embodiment, clustering the first data set includes:

[0060] S41: Acquire all data points in the first data set;

[0061] S42: Calculate the aggregation degree of each data point in the first data set, and iterate according to the aggregation degree to obtain multiple cluster centers; the data points corresponding to the cluster centers are the optional relay nodes.

[0062] Specifically, the degree of aggregation is determined by a preference parameter. The larger the preference parameter, the stronger the degree of aggregation. The calculation formula of the preference parameter is:

[0063] p=median(s(k,i)),k,i=1,2,…,N

[0064] Among them, s(k,i) represents the data point x k With data point x i The similarity between the data points x kPreference parameters.

[0065] Before the first round of iteration, the initial value of the preference parameter of each data point needs to be set to the same value, and the value is not 0; after calculating the preference parameters of each data point, the preference parameters can be sorted and the middle value of the preference parameters can be taken to assign the initial value, so that all data points are set with the same preference parameters, and then iteration is performed according to the initial value.

[0066] By inputting the sampled data set into the VAE model, the encoded data set and the decoded data set are obtained. The specific process is as follows:

[0067] The bionic cluster analysis (AP) method is used to obtain a set of candidate relay nodes. For all relay nodes x(i), i = 1, 2, ..., N, the AP algorithm uses a similarity formula to select the similarity between relay nodes as input. The similarity formula is shown in formula (3):

[0068] s(i,k)=-((x i -x k ).^2) (3)

[0069] Where s(i,k) represents the relay node x k With node x i The similarity between them; s(k,k) represents the data point x k The ability to become a cluster center cannot be set to 0. At this time, the preference parameter p is used to replace s(k, k). Before the AP algorithm starts to iterate, all data points are considered to have the same ability to become a cluster center; therefore, all data points are set with the same preference parameter, and the preference parameter selects the data point x k The median of all similarities, that is, p = median(s(k,i)), i = 1, 2, ..., N; the larger p is, the smaller the similarity is. k The stronger the ability to become a cluster center, the larger the final number of clusters; conversely, the smaller the number of clusters.

[0070] In another embodiment, in S5, importance sampling is performed in the set of candidate relay nodes, and the steps include:

[0071] S51: sampling a set of candidate relay nodes according to a preset node distribution to generate a distribution sample set;

[0072] S52: Calculate the mathematical expectation of each random variable in the distribution sample set;

[0073] S53: extracting a random variable with the largest mathematical expectation, and taking the sampling point corresponding to the random variable as the optimal relay node.

[0074] The specific process is as follows: Select the nodes in the sampling distribution that best capture the characteristics of the target distribution. The basic idea is to select nodes in the sampling distribution so that the variance of their importance weights is minimized.

[0075] Since the target distribution characteristic function f(x) itself is defined on a distribution, which is defined as π(x), it is impossible to sample directly from π(x). Therefore, a new distribution p(x) is found for sampling, and the expectation of f(x) under the distribution π(x) is indirectly obtained. The expectation of function f(x) under the probability distribution π(x) is: E[f] = ∫ x π(x)f(x)dx, this expectation cannot be obtained directly, so it is necessary to use p(x) for sampling. By sampling {x 1 ,x 2 ,…,x i ,…,x N}, and calculate the value of the function f(X) at each sample point {f(x 1 ),f(x 2 ),…,f(x i ),…,f(x N )}, and then calculate the sample mean of these function values ​​to estimate the mathematical expectation of the random variable f(X) under the p(x) distribution.

[0076] Since this distribution is not suitable for sampling and is prone to cause large deviations, the formula is rewritten as shown in formula (4):

[0077]

[0078] The expectation is as shown in formula (5):

[0079]

[0080] in, is the importance sampling correction factor.

[0081] By sampling {x 1 ,x 2 ,…,x i ,…,x N}, and calculate the value of the function f(X) at each sample point {f(x 1 ),f(x 2 ),…,f(x i ),…,f(x N )}, and then calculate the sample mean of these function values ​​to estimate the mathematical expectation of the random variable f(X) under the p(x) distribution.

[0082] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0083] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A wireless communication network relay selection method based on variational autoencoder, characterized in that: The following steps are involved: Receive wireless communication network signals; Preprocessing the wireless communication network signal to obtain initial data; Inputting the initial data into a pre-built variational autoencoder model to obtain an encoded first data set; Obtaining a set of candidate relay nodes by clustering the first data set; Obtain all data points in the first data set; The aggregation degree of each data point in the first data set is calculated, and it is iterated according to the aggregation degree to obtain multiple cluster centers; the data points corresponding to the cluster centers are the optional relay nodes; the aggregation degree is determined by a preference parameter, and the larger the preference parameter, the stronger the aggregation degree; the calculation formula of the preference parameter is: s(i,k)=-[(x i -x k ).^2] Where s(i,k) represents the data point x i With data point x k The degree of similarity between Importance sampling is performed in the candidate relay node set to obtain the optimal relay node; the steps include: Sampling the candidate relay node set according to a preset node distribution to generate a distribution sample set; Calculating the mathematical expectation of each random variable in the distribution sample set; The random variable with the largest mathematical expectation is extracted, and the sampling point corresponding to the random variable is used as the optimal relay node.

2. A wireless communication network relay selection method based on variational autoencoder according to claim 1, characterized in that: The importance sampling is specifically: One or more evaluation indicators are confirmed for comprehensive evaluation; the evaluation indicators include bit error rate and / or signal-to-noise ratio.

3. A wireless communication network relay selection method based on variational autoencoder according to claim 1, characterized in that: The training steps of the variational autoencoder model are: For the training data set X = {x1,…,x N }, the goal is to maximize the marginal likelihood: Due to the non-optimizability of marginal likelihood, when the model is parameterized using a neural network, an inference model is introduced to optimize the lower limit of the variable: lnp(x)≥E q(z|x) [lnp(x|z)]-KL(q(z|x)||p(z)) (2) Where x represents the input data point, z represents the latent variable, p(x|z) is the decoder, and p(z)=N(z|0,1) is the prior; E q(z|x) It means to find the expectation under the posterior distribution q(z|x); for continuous z, it is achieved by reparameterizing the posterior distribution q(z|x).

4. A wireless communication network relay selection method based on variational autoencoder according to claim 1, characterized in that: Before iteration, the same preference parameters are set for each of the data points.

5. The method for selecting a wireless communication network relay based on a variational autoencoder according to claim 1, characterized in that: The preset node distribution is: Among them, p(x) is the preset node distribution, π(x) is the target distribution of the node, and f(x) is the distribution characteristic function of the distribution π(x).

6. A wireless communication network relay selection method based on variational autoencoder according to claim 5, characterized in that: The mathematical expectation is calculated as follows: in, is the importance sampling correction factor.

Images