CN115642972A - Dynamic channel communication detection method, device, equipment and readable storage medium - Google Patents

Abstract

The present application discloses a dynamic channel communication detection method, device, equipment and readable storage medium. The method includes the steps of: obtaining communication data of a dynamic channel; inputting the communication data into a detection network model to obtain an output result, the The detection network model is based on the dynamic change information of the dynamic channel, and is obtained by adjusting the calculation parameters of each neuron to adapt to the change of the dynamic channel. The detection network model is based on the preset communication data training samples and the obtained The sample label of the preset communication data training sample is obtained after iteratively training the preset deep neural network model; according to the output result, the detection result of the communication data is obtained by calculation. The present application realizes that the detection network model adapts to the change of the dynamic channel, thereby detecting the dynamic channel and improving the detection performance of the dynamic communication channel.

Description

Dynamic channel communication detection method, device, equipment and readable storage medium

technical field

The present application relates to the field of communication technology, and in particular to a dynamic channel communication detection method, device, equipment and readable storage medium.

Background technique

The wireless broadband communication channel in the sea sports scene, due to the dynamic characteristics of the channel caused by the dynamic change of the maritime communication environment, from the time point of view, it mainly shows time-varying characteristics; from the point of view of the statistical model, it usually shows the statistical non-stationarity of the channel, That is, the channel cannot be modeled with a single statistical model, which makes the traditional modeling method based on the mathematical statistical model inapplicable, which in turn leads to poor performance of the conventional detection method at the communication receiving end.

Contents of the invention

In view of this, the present application provides a dynamic channel communication detection method, device, device and readable storage medium, aiming at improving the detection performance of dynamic communication channels.

In order to achieve the above purpose, the present application provides a dynamic channel communication detection method, the dynamic channel communication detection method includes the following steps:

Obtain the communication data of the dynamic channel;

Input the communication data into the detection network model to obtain an output result, the detection network model is based on the dynamic change information of the dynamic channel, and is obtained after adjusting the calculation parameters of each neuron to adapt to the change of the dynamic channel model, the detection network model is obtained after performing iterative training on a preset deep neural network model based on preset communication data training samples and sample labels of the preset communication data training samples;

According to the output result, the detection result of the communication data is obtained by calculation.

Exemplarily, before inputting the communication data into the detection network model and obtaining an output result, it includes:

Obtain dynamic change information of the dynamic channel;

The dynamic change information is input into the memory network model; wherein, the memory network model is based on the preset change information training sample and the sample label of the preset change information training sample, and iterates the preset memory neural network model obtained after training;

calculating dynamic control parameters according to the memory network model and the dynamic change information;

According to the dynamic control parameters, the calculation parameters of each neuron of the detection network model are adjusted.

Exemplarily, the calculation of dynamic control parameters according to the memory network model and the dynamic change information includes:

Obtain the hidden state parameters of the previous moment of the memory network model;

generating a dynamically changing eigenvector of the dynamic channel according to the hidden state parameter at the last moment and the dynamic change information;

Calculate the cell state parameters of the memory network model at the current moment according to the feature vector;

calculating the hidden state parameters of the memory network model at the current moment according to the cell state parameters at the current moment;

A dynamic control parameter is calculated according to the cell state parameter at the current moment and the hidden state parameter at the current moment.

Exemplarily, the calculation of the cell state parameters of the memory network model at the current moment according to the feature vector includes:

Obtain the number of hidden layers of the detection network model;

selecting preset calculation parameters for said quantity;

According to the preset calculation parameters and the eigenvectors, calculate the number of cell state parameters of the memory network model at the current moment, so that the memory network model can calculate the number of dynamic control parameters, and use the memory network model to calculate the number of dynamic control parameters. The above-mentioned number of dynamic control parameters corresponds to controlling all hidden layers of the detection network model.

Exemplarily, the adjusting calculation parameters of each neuron of the detection network model according to the dynamic control parameters includes:

Obtain the mapping relationship between the dynamic control parameters and the hidden layer of the detection network model;

According to the mapping relationship, input the number of dynamic control parameters into corresponding hidden layers in the detection network model;

Adjusting the calculation parameters of each neuron in each hidden layer; wherein, the calculation parameters of all neurons in any hidden layer are the same.

Exemplarily, before inputting the communication data into the detection network model and obtaining an output result, it includes:

performing real number processing on the communication data;

Selecting data of a preset length from the communication data after realization processing, and inputting the data of a preset length into the detection network model.

Exemplarily, the calculating and obtaining the detection result of the communication data according to the output result includes:

converting said output result into a matrix consisting of probability estimates for corresponding symbols;

calculating the maximum argument point set of said matrix;

A detection result of the symbol of the communication data is determined according to the maximum value argument point set.

Exemplarily, in order to achieve the above purpose, the present application also provides a dynamic channel communication detection device, the device includes:

An acquisition module, configured to acquire communication data of a dynamic channel;

The input module is used to input the communication data into the detection network model to obtain an output result. The detection network model is based on the dynamic change information of the dynamic channel and is obtained after adjusting the calculation parameters of each neuron to adapt to the dynamic A model of channel changes, the detection network model is obtained after iteratively training a preset deep neural network model based on preset communication data training samples and sample labels of the preset communication data training samples;

The calculation module calculates the detection result of the communication data according to the output result.

Exemplarily, in order to achieve the above purpose, the present application also provides a dynamic channel communication detection device, which includes: a memory, a processor, and a dynamic channel communication system stored in the memory and operable on the processor. A detection program, the dynamic channel communication detection program is configured to implement the steps of the above dynamic channel communication detection method.

Exemplarily, in order to achieve the above purpose, the present application also provides a computer-readable storage medium, where a dynamic channel communication detection program is stored on the computer-readable storage medium, and the dynamic channel communication detection program is implemented when executed by a processor. The steps of the above dynamic channel communication detection method.

Compared with the existing technology, the wireless broadband communication channel in the sea sports scene changes dynamically, and the channel cannot be modeled with a single statistical model, which makes the traditional mathematical statistical model method inapplicable, which in turn leads to the detection of the conventional communication receiving end. Compared with the poor performance of the method, in this application, the communication data of the dynamic channel is obtained, and the communication data of the dynamic channel is input into the detection network model to obtain the output result, wherein the detection network model is based on The dynamic change information of the dynamic channel is a model adapted to the change of the dynamic channel obtained after adjusting the calculation parameters of each neuron, wherein the detection network model is based on preset communication data training samples and the preset communication The sample label of the data training sample is obtained after iterative training of the preset deep neural network model, thus ensuring that the detection network model can adapt to the dynamic change information of the dynamic channel, thereby ensuring that the detection network model The accuracy of the output results, so that according to the output results, the detection results of the communication data are calculated, that is, the calculation parameters of the detection network model are adjusted, so that the detection network model can adapt to the dynamic changes of the dynamic channel, and the detection network model can be accurate. Correlative detection is performed on the communication data of the dynamic channel, and the accuracy of the output result is ensured, thereby ensuring the accuracy of the detection result of the communication data calculated according to the output result, thereby improving the detection performance of the dynamic communication channel.

Description of drawings

FIG. 1 is a schematic flow diagram of the first embodiment of the dynamic channel communication detection method of the present application;

FIG. 2 is a schematic flow diagram of the second embodiment of the dynamic channel communication detection method of the present application;

Fig. 3 is the network structural diagram of memory network model;

Fig. 4 is a schematic diagram of the architecture of the memory network model and the detection network model;

Fig. 5 is a schematic structural diagram of a detection network model based on dynamic control information;

FIG. 6 is a schematic structural diagram of the hardware operating environment involved in the solution of the embodiment of the present application.

The realization, functional features and advantages of the present application will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

Detailed ways

It should be understood that the specific embodiments described here are only used to explain the present application, and are not intended to limit the present application.

The present application provides a dynamic channel communication detection method. Referring to FIG. 1 , FIG. 1 is a schematic flowchart of a first embodiment of the dynamic channel communication detection method of the present application.

The embodiment of the present application provides an embodiment of a dynamic channel communication detection method. It should be noted that although the logic sequence is shown in the flow chart, in some cases, the sequence shown here may be executed in a different order. or the steps described. For the convenience of description, the following omits the execution subject to describe the various steps of the dynamic channel communication detection method. The dynamic channel communication detection method includes:

Step S110: Obtain communication data of the dynamic channel;

For the wireless broadband communication channel in the sea sports scene, in this scene, the communication transmitter of the ship in motion usually transmits the relevant communication data to the communication receiver of the base station on the shore to complete the communication process.

The communication channel between the ship's communication transmitter and the shore base station's communication receiver is a dynamic channel, and the main dynamic change of the dynamic channel is the dynamic change of the ship at sea, for example, the position of the ship is constantly changing when sailing at sea , or the communication channel is affected by the wind and waves at sea.

，调制阶数为M。 Obtain the communication data of the dynamic channel, which is the interactive data from the communication transmitter of the ship at sea to the communication receiver of the shore base station. The signal symbol set of the communication data in the initialization communication transmitter is

, and the modulation order is M.

Among them, the symbol rate can be 20M, using QPSK (Quadrature Phase Shift Keying, quadrature phase shift keying) modulation, the modulation order is M, the modulation order is the same as the number of output layers of the detection network model, and M=4 is desirable .

Wherein, the initialization is to initialize the parameters of the communication system between the ship and the base station to reach an enabled state.

Step S120: Input the communication data into the detection network model to obtain an output result. The detection network model is based on the dynamic change information of the dynamic channel and is obtained after adjusting the calculation parameters of each neuron to adapt to the dynamic channel. The model of the change situation, the detection network model is obtained after iteratively training the preset deep neural network model based on the preset communication data training samples and the sample labels of the preset communication data training samples;

In view of the problems existing in the symbol detection method of the conventional communication receiving end, and considering that the signal detection method based on the neural network is data-driven rather than model-driven, and it has the characteristics of not relying on the statistical model of the data, the conventional detection method can be The statistical model of the data in the network is replaced by the neural network model, so as to avoid the problems existing in modeling the communication channel with a single statistical model.

A new symbol detection method at the receiving end is designed based on the deep neural network, that is, the relevant data is calculated through the deep neural network model, and the vector results output by it are further calculated to obtain the symbol detection result at the receiving end.

When replacing the conventional single statistical model with a deep neural network model, each neuron in the deep neural network model needs to be adjusted to adapt to the dynamic changes of the dynamic channel. , to adjust the calculation parameters of each neuron of the deep neural network model.

The large-scale dynamic changes of the channel in the current maritime motion scene lead to the problems of inapplicability of conventional symbol detection methods and poor communication performance. Therefore, before performing conventional symbol detection on the communication channel, it is necessary to analyze the dynamic change of the communication channel and make corresponding countermeasures.

Therefore, the dynamic change information of the dynamic channel is acquired, and the dynamic change information refers to the characteristic information of the dynamic change of the communication channel.

Exemplarily, the dynamic change information includes the position change of the communication channel and the dynamic change of the communication signal during communication.

Among them, the position change of the communication channel includes the distance between the ship and the shore base station, and the speed of the ship itself, which affects the rate of change in the distance.

Wherein, the dynamic change of the signal of the communication channel is the change of the strength of the signal. During the calculation, the influence on the calculation result due to the large fluctuation of the signal is avoided. Therefore, the average strength of the overall signal is used for the change of the strength.

Among them, the calculation process of the neuron itself has various functions and parameters of the neuron used for calculation, and according to the dynamic change information, continuously adjusting the calculation parameters of each neuron can affect the output after calculation, so that the depth The neural network model adapts to the dynamic changes of the dynamic channel.

Among them, after the iterative training of the deep neural network model through the preset communication data training samples and the sample labels of the preset communication data training samples, a detection network model is obtained, which is mainly used to detect relevant The data is calculated by the neural network model.

The above-mentioned preset communication data training samples and the sample labels of the preset communication data training samples are determined according to the content that needs to be processed when actually calculating communication-related data. The training sample labels, gradient calculation and other processes are commonly used neural network model training methods, so I won’t repeat them here.

The following detection network models are all trained deep neural network models.

Exemplarily, the network of the detection network model has a total of L+2 layers.

。该接收信号y即为基站的通信接收端通过 动态信道而接收到的信号，即通信数据。 Among them, the input layer is the received signal after real digitization

. The received signal y is a signal received by the communication receiving end of the base station through a dynamic channel, that is, communication data.

。 Among them, the hidden layer has L layers, and the number of neuron nodes corresponding to each hidden layer is

.

。 The output layer of the deep neural network is a fully connected layer, and its output is calculated as

.

The above L-layer network can determine the specific number of layers according to actual needs.

Step S150: Calculate and obtain the detection result of the communication data according to the output result.

According to the above content, after the detection network model outputs the corresponding calculation results, the symbol detection of the communication data can be completed by using the conventional symbol detection method at the receiving end.

The detection result of the communication data is the symbol detection of the communication data.

Exemplarily, before inputting the communication data into the detection network model and obtaining an output result, it includes:

Step a: performing real number processing on the communication data;

Because the input layer of the detection network model needs to receive the received signal after real digitization, it is necessary to process the communication data into real digits before inputting the communication data into the detection network model.

When performing real number processing on communication data, use the real number function:

When the realization function is used, the real part and the imaginary part of the communication data are respectively extracted, and the real part and the imaginary part are composed into a communication vector.

。 At this time, the real-numbered communication data is expressed as:

.

Step c: selecting data of a preset length from the communication data processed by real digitization, and inputting the data of a preset length into the detection network model.

，选取相应数量个符号作为输入数据序列。 According to the communication vector, that is, according to

, select the corresponding number of symbols as the input data sequence.

个符号作为检测网络模型的输入数据序列，其序列长度为

，即为预设长 度。 Wherein, the corresponding number is determined according to the communication vector, that is, before each calculation of the detection network model, the adjacent

symbols as the input data sequence of the detection network model, and its sequence length is

, which is the preset length.

Exemplarily, the calculating and obtaining the detection result of the communication data according to the output result includes:

Step d: converting the output result into a matrix composed of probability estimates of corresponding symbols;

，其中M为通信系统符号的调制阶数，

的具体矩 阵表达式如下： The output of the deep neural network

, where M is the modulation order of the symbol in the communication system,

The specific matrix expression of is as follows:

为相应符号的概率估计值。 That is, each element corresponds to the current symbol to be detected

is the probability estimate for the corresponding symbol.

Step e: calculating the maximum independent variable point set of the matrix;

Step f: Determine the detection result of the symbol of the communication data according to the point set of the maximum value argument.

The symbol detection method of the conventional receiving end can complete the symbol detection of the communication data. The detection process calculates the maximum independent variable point set in the above matrix. At this time, the relevant general calculation formula can be used. Therefore, it can be calculated by the following formula Get the maximum independent variable point set:

即为检测结果的向量大小。 in,

That is, the vector size of the detection result.

After the point set of the maximum value argument is calculated, the detection result of the symbol of the communication data is determined, that is, the maximum value is taken as the detection result.

Compared with the existing technology, the wireless broadband communication channel in the sea sports scene changes dynamically, and the channel cannot be modeled with a single statistical model, which makes the traditional mathematical statistical model method inapplicable, which in turn leads to the detection of the conventional communication receiving end. Compared with the poor performance of the method, in this application, the communication data of the dynamic channel is obtained, and the communication data of the dynamic channel is input into the detection network model to obtain the output result, wherein the detection network model is based on The dynamic change information of the dynamic channel is a model adapted to the change of the dynamic channel obtained after adjusting the calculation parameters of each neuron, wherein the detection network model is based on preset communication data training samples and the preset communication The sample label of the data training sample is obtained after iterative training of the preset deep neural network model, thus ensuring that the detection network model can adapt to the dynamic change information of the dynamic channel, thereby ensuring that the detection network model The accuracy of the output results, so that according to the output results, the detection results of the communication data are calculated, that is, the calculation parameters of the detection network model are adjusted, so that the detection network model can adapt to the dynamic changes of the dynamic channel, and the detection network model can be accurate. Correlative detection is performed on the communication data of the dynamic channel, and the accuracy of the output result is ensured, thereby ensuring the accuracy of the detection result of the communication data calculated according to the output result, thereby improving the detection performance of the dynamic communication channel.

For example, refer to FIG. 2, which is a schematic flowchart of the second embodiment of the dynamic channel communication detection method of the present application. Based on the above-mentioned first embodiment of the dynamic channel communication detection method of the present application, a second embodiment is proposed. The method also includes include:

Step S210: Obtain dynamic change information of the dynamic channel;

The large-scale dynamic changes of the channel in the current maritime motion scene lead to the problems of inapplicability of conventional symbol detection methods and poor communication performance. Therefore, before performing conventional symbol detection on the communication channel, it is necessary to analyze the dynamic change of the communication channel and make corresponding countermeasures.

Therefore, the dynamic change information of the dynamic channel is acquired, and the dynamic change information refers to the characteristic information of the dynamic change of the communication channel.

Exemplarily, the dynamic change information includes the position change of the communication channel and the dynamic change of the communication signal during communication.

Among them, the position change of the communication channel includes the distance between the ship and the shore base station, and the speed of the ship itself, which affects the rate of change in the distance.

Wherein, the dynamic change of the signal of the communication channel is the change of the strength of the signal. During the calculation, the influence on the calculation result due to the large fluctuation of the signal is avoided. Therefore, the average strength of the overall signal is used for the change of the strength.

Step S220: Input the dynamic change information into the memory network model; wherein, the memory network model is based on the preset change information training sample and the sample label of the preset change information training sample, for the preset memory neural network The model is obtained after iterative training;

When adjusting the calculation parameters of each neuron in the detection network model according to the dynamic change information, it is necessary to process the dynamic change information first. This process takes into account that the dynamic change process of the dynamic channel has a certain correlation, so the memory neural network is used Realize the learning and memory of the dynamic environment, record its dynamic change process continuously, and calculate the state of the dynamic channel at the current moment according to the state of the dynamic channel at the previous moment and the dynamic change information.

In the above calculation process, the memory state parameters of the memory neural network are used to memorize and learn the dynamic change information of the communication channel, that is, to memorize the position and calculation of the ship at the previous moment mainly based on the distance, speed and average signal strength in the dynamic change information. The position of the ship at the current moment, so as to determine the change of the dynamic channel.

The memory network model is the trained memory neural network model, and the preset memory neural network model is iteratively trained using the preset change information training samples and the sample labels of the preset change information training samples to memory the network model.

The preset change information training sample is composed of dynamic change information, and the sample label of the preset change information training sample is determined according to the calculation result of the dynamic change information. The training process of the memory network model is similar to the training process of the detection network model, and the training The process is a common technology, and will not be repeated here.

Step S220: Calculate and obtain dynamic control parameters according to the memory network model and the dynamic change information;

The memory network model after the training is used to calculate the dynamic change information, thereby calculating the dynamic control parameters, where the dynamic control parameters are used as the control parameters to adjust the calculation parameters in the detection network model, so that the detection network model adapts to the dynamic channel changes.

That is, the dynamic control parameter is threshold control information for controlling the detection network model. Therefore, the length of the dynamic control parameter is consistent with the number of hidden layers in the detection network model, that is, the length of the dynamic control parameter is L. In this way, the threshold control of the hidden layer of the detection network model is achieved.

Exemplarily, the calculation of dynamic control parameters according to the memory network model and the dynamic change information includes:

Step g: Obtain the hidden state parameters of the memory network model at the previous moment;

After each dynamic change information is input into the memory network model, the network model will update the memory state parameters in the neural network, thereby memorizing the current dynamic change information and related calculation of the dynamic change information, so as to calculate the dynamic Control parameters.

， 以及该神经网络的细胞状态参数

，在记忆网络模型每次接到动态变化信息时，均会对上 述两个记忆状态参数进行更新。 The memory state parameters in the memory network model include the hidden state parameters of the neural network in the memory network model

, and the cell state parameters of the neural network

, each time the memory network model receives dynamic change information, it will update the above two memory state parameters.

和

为记忆网络模型当前时刻t的记忆状态参数，其向量长度均为

；其 中，可根据隐藏状态参数计算出细胞状态参数，其中，隐藏状态参数的初始值可选为1。 in,

and

is the memory state parameter of the memory network model at the current moment t, and its vector length is

;wherein, the cell state parameters can be calculated according to the hidden state parameters, wherein the initial value of the hidden state parameters can be selected as 1.

When calculating the cell state parameters through the hidden state parameters, it is necessary to use the hidden state parameters of the previous moment and the dynamic change information of the current moment, that is, the hidden state parameters of the previous moment updated according to the dynamic change information of the previous moment input into the memory network model. State parameters, calculate the hidden state parameters at the current moment.

Therefore, the hidden state parameters of the last moment in the memory network model are obtained.

，上一时刻的隐藏状态参数记作

。The hidden state parameters of the memory network model at the current moment are denoted as

, the hidden state parameter at the last moment is denoted as

.

Step h: generating a dynamically changing eigenvector of the dynamic channel according to the hidden state parameters at the last moment and the dynamic change information;

The calculation process in the neural network model usually uses vector calculation. Therefore, the hidden state parameters and dynamic change information of the previous moment can be used to generate a new feature vector about the dynamic change of the dynamic channel, which includes the characteristics of the previous moment, and Dynamic change information at the current moment.

From the above, it can be known that the dynamic change information includes distance, speed and average signal strength, which are respectively recorded as distance d, speed v and average signal strength r.

。 Therefore, the generated eigenvectors are written as:

.

Step i: Calculate the cell state parameters of the memory network model at the current moment according to the feature vector;

The output of the memory network model requires not only the hidden state parameters but also the cell state parameters, which can be calculated based on the eigenvectors.

In the calculation process, it is also necessary to obtain the network parameters of the memory network model, which are specifically the parameters of each neural node in the memory network model.

The specific calculation formula is as follows:

为神经网络里的逐点逻辑激活函数； in,

is the point-by-point logical activation function in the neural network;

tanh is a point-wise hyperbolic tangent activation function;

为遗忘门限参数；

is the forgetting threshold parameter;

为输入门限参数；

is the input threshold parameter;

为输出门限参数；

is the output threshold parameter;

为临时细胞状态参数；

is the temporary cell state parameter;

为记忆网络模型的上一时刻的细胞状态参数；

is the cell state parameter at the last moment of the memory network model;

为特征向量；

is the feature vector;

和

为记忆网络模型的各神经节点参数；

and

are the parameters of each neural node of the memory network model;

和

为

的偏差向量。

and

for

deviation vector.

Step m: Calculate the hidden state parameters of the memory network model at the current moment according to the cell state parameters at the current moment;

和动态变化信息，计算出

，再根据

，计算得到

。 After calculating the cell state parameters at the current moment, calculate the hidden state parameters of the memory network model at the current moment according to the cell state parameters, that is, according to

and dynamic change information, calculate the

, and then according to

, calculated to get

.

的计算公式如下所示：

The calculation formula of is as follows:

为输出门限参数； in,

is the output threshold parameter;

tanh is a point-wise hyperbolic tangent activation function;

Step n: Calculate and obtain dynamic control parameters according to the cell state parameters at the current moment and the hidden state parameters at the current moment.

。 After calculating the cell state parameters at the current moment and the hidden state parameters at the current moment, the dynamic control parameters at the current moment are calculated, and the dynamic control parameters are denoted as

.

的计算公式如下所示：

The calculation formula of is as follows:

是逐点软激活函数，它的第i个元素

为

； in,

Is a point-wise soft activation function, its i-th element

for

;

是

参数权重矩阵；

yes

parameter weight matrix;

是3

1的偏差向量。

is 3

A bias vector of 1.

Referring to FIG. 3 , FIG. 3 is a network structure diagram of a memory network model.

The network takes the dynamic change information as input and calculates the cell state parameters and hidden state parameters through the calculation of relevant neurons. At the same time, according to the characteristics of the memory neural network, the relevant state parameters at the previous moment are retained and used to calculate the current moment. In the relevant state parameters of .

Finally, the dynamic control information is output according to the cell state parameters at the current moment and the hidden state parameters of the current state.

Exemplarily, the calculation of the cell state parameters of the memory network model at the current moment according to the feature vector includes:

Step j: Obtain the number of hidden layers of the detection network model;

Step k: selecting the number of preset calculation parameters;

The output result of the memory network model will be input into the detection network model, and the parameters of the neurons of the detection network model will be adjusted. Therefore, the output result of the memory network model should correspond to the hidden layer of the detection network model, that is, for the detection network model. The number of hidden layers, which sets the number of parameters to be calculated by the memory network model.

Thus, get the number of hidden layers of the detection network model.

And select a corresponding number of preset calculation parameters according to the number of hidden layers, the preset calculation parameters are the neural node parameters of the memory network model, and select different preset calculation parameters according to the number of hidden layers.

和

为

的矩阵。 Exemplarily, the number of hidden layers is 3 as an example, that is, the memory network model needs to output three parameters, and the calculation parameters of the neurons of the detection network model are controlled according to the three parameters. Therefore, according to the number of hidden layers When selecting the preset calculation parameters, select the memory network model

and

for

matrix.

When the number of hidden layers is any other value, the effect of selecting the preset calculation parameters is similar to that when the number of hidden layers is 3, which will not be repeated here.

Step 1: According to the preset calculation parameters and the eigenvectors, calculate the number of cell state parameters of the memory network model at the current moment, so that the memory network model can calculate the number of dynamic control parameters, And correspondingly control all hidden layers of the detection network model through the number of dynamic control parameters.

According to the preset calculation parameters and eigenvectors, that is, according to the calculation formula given above, the cell state parameters at the current moment of the number of hidden layers can be calculated, and the cell state parameters at the current moment of the number of hidden layers can be calculated, Calculate the hidden state parameters at the current moment of the number of hidden layers, thereby calculating the dynamic control parameters of the number of hidden layers, so that the output result of the memory network model can control all hidden layers of the detection network model.

包括 三个数值，分别记作

，即

。 Exemplarily, the above-mentioned number of hidden layers is 3 as an example to continue the explanation, that is, the final calculated

Contains three values, denoted as

,Right now

.

的数量即可。 When the number of hidden layers is other values, increase or decrease accordingly

The quantity can be.

Step S230: Adjust calculation parameters of each neuron of the detection network model according to the dynamic control parameters.

中的数值与检测网络模型中的隐藏层一一对应 关系，使得

中的任一数值仅控制一个隐藏层。 After the dynamic control parameters are calculated, according to the dynamic control parameters, the calculation parameters of each neuron of the detection network model are adjusted. The adjustment process is based on

The values in are in one-to-one correspondence with the hidden layers in the detection network model, so that

Any value in controls only one hidden layer.

That is, through the memory network model, according to the dynamic change information, calculate and output the corresponding dynamic control information, input the dynamic control information into the detection network model, and control the calculation parameters in the detection network model, so that after the signal is realized, the input The final calculation result of the data in the detection network model is dynamically changed to adapt to the changing situation of the dynamic channel.

Referring to FIG. 4, FIG. 4 is a schematic diagram of the architecture of the memory network model and the detection network model.

Wherein, the memory network model and the detection network model can be used as an integrated neural network model, or as two independent neural network models.

Among them, the dynamic change information is input to the memory network model, and the dynamic control information is output through the memory network model.

。 Among them, input the real-numbered communication data y to the detection network model, output the corresponding result through the detection network model, and perform symbol detection on the result to obtain the detection result

.

Exemplarily, the adjusting calculation parameters of each neuron of the detection network model according to the dynamic control parameters includes:

Step o: Obtain the mapping relationship between the dynamic control parameters and the hidden layer of the detection network model;

中的数值与检测网络模型的隐藏层之间的对应关系，例如，在隐 藏层的数量为3时，

，检测网络模型的隐藏层的数量为3，此时将根 据

所控制检测网络模型的第一个隐藏层、

控制检测网络模型的第二个隐藏层、

控制检测网络模型的第三个隐藏层。 The mapping relationship is

The correspondence between the values in and the hidden layers of the detection network model, for example, when the number of hidden layers is 3,

, the number of hidden layers of the detection network model is 3, at this time it will be based on

The first hidden layer of the controlled detection network model,

Controlling the second hidden layer of the detection network model,

Controls the third hidden layer of the detection network model.

Step p: according to the mapping relationship, respectively input the number of dynamic control parameters into corresponding hidden layers in the detection network model;

的相应数据对应调整隐藏层的神经元的原本的计算参数，且在调整过程 中，

会控制整个隐藏层中的全部神经元的计算参数统一，均为

中的同一个值，

。 thus will

The corresponding data of corresponding to adjust the original calculation parameters of neurons in the hidden layer, and during the adjustment process,

It will control the calculation parameters of all neurons in the entire hidden layer to be unified, which are

the same value in ,

.

Step q: adjusting the calculation parameters of each neuron in each hidden layer; wherein, the calculation parameters of all neurons in any hidden layer are the same.

Exemplarily, the detection network model has a total of 5 layers as an example for illustration, where there are 1 input layer, 1 output layer, and 3 hidden layers respectively.

，每次计算将相邻的

个符 号作为输入序列，其序列长度为

. Among them, according to the above, it can be seen that the input layer is the received signal after real digitization

, each calculation will adjacent

symbols as the input sequence, the sequence length is

.

，

，

。 Among them, the number of neuron nodes corresponding to the hidden layer is

,

,

.

Referring to FIG. 5 , FIG. 5 is a schematic structural diagram of a detection network model based on dynamic control information.

。 Figure 5 takes the detection network model with three hidden layers as an example, where the leftmost is the input layer, and the real-numbered communication data y is input into the detection network model, through hidden layer 1, hidden layer 2 and hidden layer The relevant calculation of layer 3, and finally output the corresponding result in the rightmost output layer

.

，相应调整隐藏层的计算参数。 According to the structure in Figure 5, the detection network model is used to perform related calculations, wherein the hidden layer of the detection network model will be input according to the memory network model

, adjust the calculation parameters of the hidden layer accordingly.

Among them, the output of hidden layer 1 is as follows:

是

权重参数矩阵，

是

的偏差向量； in

yes

weight parameter matrix,

yes

the deviation vector;

为深度神经网络里的逐点逻辑激活函数；

为动态控制参数。

is a point-by-point logical activation function in a deep neural network;

is a dynamic control parameter.

Among them, the output of hidden layer 2 is as follows:

是

权重参数矩阵；

是

的偏差向量； in

yes

weight parameter matrix;

yes

the deviation vector;

为深度神经网络里的逐点逻辑激活函数；

为动态控制参数。

is a point-by-point logical activation function in a deep neural network;

is a dynamic control parameter.

Among them, the output of hidden layer 3 is as follows:

是

权重参数矩阵；

是

的偏差向量； in

yes

weight parameter matrix;

yes

the deviation vector;

为深度神经网络里的逐点逻辑激活函数；

为动态控制参数。

is a point-by-point logical activation function in a deep neural network;

is a dynamic control parameter.

The output layer of the deep neural network is a fully connected layer, and its output is as follows:

是

权重参数矩阵； in

yes

weight parameter matrix;

是

的偏差向量；

yes

the deviation vector;

是逐点软激活函数，它的第i个元素

为

；

Is a point-wise soft activation function, its i-th element

for

;

为调制阶数。

is the modulation order.

对符号进行相应检测，得到检测结果。 Finally, according to

Corresponding detection is carried out on the symbol, and the detection result is obtained.

In this embodiment, the memory network model is used to calculate the dynamic change information accordingly, so that the memory network model learns and remembers the dynamic change information of the dynamic channel at the previous moment, and calculates the dynamic change information of the current moment, so that the calculated The parameters are adapted to the changing situation of the dynamic channel, and the dynamic control parameters are further calculated. According to the dynamic control parameters, the calculation parameters of each neuron of the detection network model are adjusted, so that the detection network model can detect the communication data of the dynamic channel.

In addition, the present application also provides a dynamic channel communication detection device, and the dynamic channel communication detection device includes:

An acquisition module, configured to acquire communication data of a dynamic channel;

The input module is used to input the communication data into the detection network model to obtain an output result. The detection network model is based on the dynamic change information of the dynamic channel and is obtained after adjusting the calculation parameters of each neuron to adapt to the dynamic A model of channel changes, the detection network model is obtained after iteratively training a preset deep neural network model based on preset communication data training samples and sample labels of the preset communication data training samples;

A calculation module, configured to calculate the detection result of the communication data according to the output result.

Exemplarily, the input module includes:

An acquisition submodule, configured to acquire dynamic change information of the dynamic channel;

The first input submodule is used to input the dynamic change information into the memory network model; wherein, the memory network model is based on the preset change information training sample and the sample label of the preset change information training sample, and the preset change information obtained after iterative training of the memory neural network model;

The first calculation submodule is used to calculate dynamic control parameters according to the memory network model and the dynamic change information;

The adjustment sub-module is used to adjust the calculation parameters of each neuron of the detection network model according to the dynamic control parameters.

Exemplary, the calculation sub-module includes:

The calculation of dynamic control parameters according to the memory network model and the dynamic change information includes:

A first acquisition unit, configured to acquire the hidden state parameters of the memory network model at the previous moment;

A generating unit, configured to generate a dynamically changing eigenvector of the dynamic channel according to the hidden state parameters at the last moment and the dynamic change information;

A first calculation unit, configured to calculate the cell state parameters of the memory network model at the current moment according to the feature vector;

The second calculation unit is used to calculate the hidden state parameters of the memory network model at the current moment according to the cell state parameters at the current moment;

The third calculation unit is configured to calculate dynamic control parameters according to the current cell state parameters and the current hidden state parameters.

Exemplarily, the first computing unit includes:

An acquisition subunit, configured to acquire the number of hidden layers of the detection network model;

selecting subunits for selecting preset calculation parameters of the quantity;

The calculation subunit is used to calculate the current cell state parameters of the number in the memory network model according to the preset calculation parameters and the eigenvector, so that the memory network model can calculate the dynamics of the number Control parameters, and correspondingly control all hidden layers of the detection network model through the number of dynamic control parameters.

Exemplarily, the adjustment submodule includes:

A second acquisition unit, configured to acquire a mapping relationship between the dynamic control parameters and the hidden layer of the detection network model;

an input unit, configured to respectively input the number of dynamic control parameters into corresponding hidden layers in the detection network model according to the mapping relationship;

The adjustment unit is configured to adjust the calculation parameters of each neuron in each hidden layer; wherein, the calculation parameters of all neurons in any hidden layer are the same.

Exemplarily, the input module includes:

A processing submodule, configured to process the communication data in real numbers;

The second input sub-module is used to select data of a preset length from the communication data processed by realization, and input the data of a preset length to the detection network model.

Exemplarily, the calculation module includes:

A conversion submodule, configured to convert the output result into a matrix consisting of probability estimates of corresponding symbols;

The second calculation submodule is used to calculate the maximum value independent variable point set of the matrix;

The determination submodule is configured to determine the detection result of the symbol of the communication data according to the maximum value argument point set.

The specific implementation manners of the dynamic channel communication detection device of the present application are basically the same as the above embodiments of the dynamic channel communication detection method, and will not be repeated here.

In addition, the present application also provides a dynamic channel communication detection device. As shown in FIG. 6 , FIG. 6 is a schematic structural diagram of a hardware operating environment involved in the solution of the embodiment of the present application.

Exemplarily, FIG. 6 is a schematic structural diagram of a hardware operating environment of the dynamic channel communication detection device.

As shown in Figure 6, the dynamic channel communication detection device may include a processor 601, a communication interface 602, a memory 603 and a communication bus 604, wherein the processor 601, the communication interface 602 and the memory 603 complete mutual communication through the communication bus 604 , the memory 603 is used to store computer programs; the processor 601 is used to implement the steps of the dynamic channel communication detection method when executing the programs stored in the memory 603 .

The communication bus 604 mentioned above in the dynamic channel communication detection device may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (Extended Industry Standard Architecture, EISA) bus or the like. The communication bus 604 can be divided into address bus, data bus, control bus and so on. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

The communication interface 602 is used for communication between the dynamic channel communication detection device and other devices.

The memory 603 may include a random access memory (Random Access Memory, RMD), and may also include a non-volatile memory (Non-Volatile Memory, NM), such as at least one disk memory. Optionally, the memory 603 may also be at least one storage device located away from the aforementioned processor 601 .

The above-mentioned processor 601 may be a general-purpose processor, including a central processing unit (Central Processing Unit, CPU), a network processor (Network Processor, NP), etc.; it may also be a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), field programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

The specific implementation manners of the dynamic channel communication detection device of the present application are basically the same as the embodiments of the above dynamic channel communication detection method, and will not be repeated here.

In addition, the embodiment of the present application also proposes a computer-readable storage medium, where a dynamic channel communication detection program is stored on the computer-readable storage medium, and when the dynamic channel communication detection program is executed by a processor, the above-mentioned dynamic The steps of the channel communication detection method.

The specific implementation manners of the computer-readable storage medium of the present application are basically the same as the above embodiments of the dynamic channel communication detection method, and will not be repeated here.

It should be noted that, as used herein, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article or system comprising a set of elements includes not only those elements, It also includes other elements not expressly listed, or elements inherent in the process, method, article, or system. Without further limitations, an element defined by the phrase "comprising a..." does not preclude the presence of additional identical elements in the process, method, article or system comprising that element.

The serial numbers of the above embodiments of the present application are for description only, and do not represent the advantages and disadvantages of the embodiments.

Through the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus a necessary general-purpose hardware platform, and of course also by hardware, but in many cases the former is better implementation. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product in essence or the part that contributes to the prior art, and the computer software product is stored in a storage medium as described above (such as ROM/RAM , magnetic disk, optical disk), including several instructions to enable a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in various embodiments of the present application.

The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structures or equivalent process transformations made by using the description of the application and the accompanying drawings are directly or indirectly used in other related technical fields. , are all included in the patent protection scope of the present application in the same way.

Claims (10)

1. A dynamic channel communication detection method, comprising:

acquiring communication data of a dynamic channel;

inputting the communication data into a detection network model to obtain an output result, wherein the detection network model is a model which is obtained by adjusting calculation parameters of each neuron based on dynamic change information of the dynamic channel and adapts to the change condition of the dynamic channel, and the detection network model is obtained by performing iterative training on a preset deep neural network model based on a preset communication data training sample and a sample label of the preset communication data training sample;

and calculating to obtain the detection result of the communication data according to the output result.

2. The dynamic channel communication detection method as claimed in claim 1, wherein said inputting said communication data to a detection network model to obtain an output result comprises:

acquiring dynamic change information of the dynamic channel;

inputting the dynamic change information into a memory network model; the memory network model is obtained by performing iterative training on a preset memory neural network model based on a preset change information training sample and a sample label of the preset change information training sample;

calculating to obtain dynamic control parameters according to the memory network model and the dynamic change information;

and adjusting the calculation parameters of each neuron of the detection network model according to the dynamic control parameters.

3. The dynamic channel communication detection method as claimed in claim 2, wherein said calculating a dynamic control parameter according to the memory network model and the dynamic variation information comprises:

acquiring a hidden state parameter of the memory network model at the last moment;

generating a dynamically changed feature vector of the dynamic channel according to the hidden state parameter at the previous moment and the dynamically changed information;

calculating the cell state parameters of the memory network model at the current moment according to the feature vectors;

calculating hidden state parameters of the memory network model at the current moment according to the cell state parameters at the current moment;

and calculating to obtain dynamic control parameters according to the cell state parameters at the current moment and the hidden state parameters at the current moment.

4. The dynamic channel traffic detection method of claim 3, wherein said calculating the cell state parameter of the memory network model at the current time according to the feature vector comprises:

acquiring the number of hidden layers of the detection network model;

selecting the number of preset calculation parameters;

and calculating the cell state parameters of the memory network model at the current moment of the number according to the preset calculation parameters and the feature vectors so as to obtain the dynamic control parameters of the number by calculation of the memory network model, and correspondingly controlling all hidden layers of the detection network model through the dynamic control parameters of the number.

5. The dynamic channel communication detection method of claim 4, wherein said adjusting the calculation parameters of each neuron of the detection network model according to the dynamic control parameters comprises:

acquiring a mapping relation between the dynamic control parameters and a hidden layer of the detection network model;

according to the mapping relation, the dynamic control parameters of the quantity are respectively input to the corresponding hidden layers in the detection network model;

adjusting the calculation parameters of each neuron of each hidden layer; wherein, the calculation parameters of all the neurons in any hidden layer are the same.

6. The dynamic channel communication detection method as claimed in claim 1, wherein said inputting said communication data to a detection network model to obtain an output result comprises:

performing real quantization processing on the communication data;

and selecting data with preset length from the communication data after real quantization processing, and inputting the data with the preset length into the detection network model.

7. The dynamic channel communication detection method as claimed in claim 6, wherein said calculating a detection result of said communication data according to said output result comprises:

converting the output result into a matrix consisting of probability estimation values of corresponding symbols;

calculating a maximum independent variable point set of the matrix;

and determining a detection result of the symbol of the communication data according to the maximum independent variable point set.

8. A dynamic channel communication detection apparatus, comprising:

the acquisition module is used for acquiring communication data of the dynamic channel;

the input module is used for inputting the communication data into a detection network model to obtain an output result, the detection network model is a model which is adaptive to the change condition of the dynamic channel and is obtained by adjusting the calculation parameters of each neuron based on the dynamic change information of the dynamic channel, and the detection network model is obtained by carrying out iterative training on a preset deep neural network model based on a preset communication data training sample and a sample label of the preset communication data training sample;

and the calculation module calculates to obtain the detection result of the communication data according to the output result.

9. A dynamic channel communication detection device, the device comprising: a memory, a processor, and a dynamic channel communication detection program stored on the memory and executable on the processor, the dynamic channel communication detection program configured to implement the steps of the dynamic channel communication detection method of any of claims 1 to 7.

10. A computer-readable storage medium, having a dynamic channel communication detection program stored thereon, which when executed by a processor implements the steps of the dynamic channel communication detection method of any one of claims 1 to 7.

Images