阅读说明：本技术 基于边缘计算和离散随机卷积的数据包合法性确认方法 (Data packet legitimacy confirmation method based on edge calculations and Discrete Stochastic convolution ) 是由 谢非佚 许爱东 文红 蒋屹新 张宇南 徐鑫辰 于 2019-09-04 设计创作，主要内容包括：本发明公开了一种基于边缘计算和离散随机卷积的数据包合法性确认方法,包括以下步骤：在边缘服务器与已知终端设备中预存原始导频信号；已知终端设备在发送信号中插入已知的原始导频信号发送给边缘服务器；边缘服务器对接收信号进行导频分离,得到接收到的导频信号矩阵；计算信道矩阵的估计值；对于已知终端设备,测得多个信道矩阵的估计值的集合；对于不同的已知终端设备,测得对应的信道矩阵估计的集合,构建训练集合；建立卷积核与卷积核移动规则,训练得到成熟的神经网络分类器；测得待验证终端设备信道矩阵估计的集合；对待验证终端设备的信道矩阵进行分类。本发明提高了卷积神经网络构建的分类器在MIMO信道矩阵中识别效果,提高识别的准确率。(The invention discloses a kind of data packet legitimacy confirmation method based on edge calculations and Discrete Stochastic convolution, comprising the following steps: prestore original pilot signal in Edge Server and known terminal equipment；Known terminal equipment is inserted into known original pilot signal in sending signal and is sent to Edge Server；Edge Server carries out pilot tone separation, the pilot signal matrix received to signal is received；Calculate the estimated value of channel matrix；For known terminal equipment, the set of the estimated value of multiple channel matrixes is measured；For different known terminal equipment, the set of corresponding channel matrices estimation is measured, constructs training set；Convolution kernel and convolution kernel movement rule are established, training obtains mature neural network classifier；Measure the set of terminal device channel matrices estimation to be verified；Classify to the channel matrix of terminal device to be verified.The present invention improve convolutional neural networks building classifier in mimo channel matrix recognition effect, improve the accuracy rate of identification.)

1. the data packet legitimacy confirmation method based on edge calculations and Discrete Stochastic convolution, it is characterised in that: including following step It is rapid:

S1. unified original pilot signal X (p) is prestored in Edge Server and multiple known terminal equipment；

S2. after known terminal equipment is inserted into known original pilot signal according to setting rule in the data packet for sending signal, Edge Server is sent to using single antenna or multiple antennas；

S3. Edge Server receives the data packet that known terminal equipment is sent using multiple antennas receiving array, and to each antenna The signal received carries out pilot tone separation, the pilot signal matrix received:

Y (p)=[Y₁(p),Y₂(p),…,Y_K(p)]；

S4. Edge Server calculates the estimated value of channel matrix

Wherein, X (p)^-1Represent the inverse matrix of X (p)；

S5. it for any known terminal equipment, repeats step S2~S4 and carries out multiple data sampling and channel matrices estimation, obtain The set of the estimated value of multiple channel matrixes:

Wherein,Represent the channel matrix obtained to m-th of known terminal equipment n-th sampled- data estimation；N=1,2, 3 ..., N, N indicate the number that data acquisition and channel matrices estimation are carried out to terminal device；

S6. for different known terminal equipment, step S2~S5 is repeated, the channel matrix for obtaining each known terminal equipment is estimated The set S of meter₁,S₂,…,S_M, wherein S_mRepresent the set of the channel matrices estimation of m-th of equipment, m=1,2,3 ..., M；

S7. according to the channel matrices estimation set S of each known terminal equipment₁,S₂,…,S_M, construct training set

S8. establish convolution kernel and convolution kernel movement rule: convolution kernel is continuous translation in longitudinal movement, be laterally it is discrete with Machine bounce:

S801. establishing convolution kernel is M ' × N ' matrix, expression are as follows:

CK=[L₁,L₂,…,L_N′]

Wherein L_n′The column vector for being M ' for dimension, M ' < M, N ' < N；

S802. convolution kernel longitudinal movement rule is identical as traditional convolutional neural networks, is continuous translation, and step-length is set as 1；

S803. convolution kernel transverse shifting rule is Discrete Stochastic bounce, specific as follows:

Convolution kernel initial position is CK₀=[L₁,L₂,…,L_N′], it is CK after the translation that traditional step-length is 1₁'=[L₂,L₃,…, L_N′+1], it will be appreciated that delete first row, and add last column right-hand column；Discrete Stochastic bounce rule are as follows: first row is deleted, And the random column in last addition in addition to previous step convolution kernel includes column, it may be assumed that

S9. according to the convolution kernel and convolution kernel movement rule established in S8, neural network classification is constructed using convolutional neural networks Device, and classifier is trained using the training set in step S7, obtain mature neural network classifier；

S10. to terminal device to be verified, step S2~S5 is repeated, the collection of terminal device channel matrices estimation to be verified is obtained Close S_m′；Classified using mature neural network classifier to the channel matrix of terminal device to be verified, judges whether it closes Method, to realize the data packet legitimacy confirmation of the equipment.

2. the data packet legitimacy confirmation method according to claim 1 based on edge calculations and Discrete Stochastic convolution, Be characterized in that: the pilot signal X (p) is one-dimensional column matrix.

3. the data packet legitimacy confirmation method according to claim 1 based on edge calculations and Discrete Stochastic convolution, Be characterized in that: the step S7 includes:

For the channel matrix set of each known terminal equipment, it is feedback with corresponding terminal device number, obtains training set It closesWherein, the corresponding training set of the channel matrix set of m-th of known terminal equipmentAre as follows:

4. the data packet legitimacy confirmation method according to claim 3 based on edge calculations and Discrete Stochastic convolution, It is characterized in that: in the step S9, with each of the channel matrix set of each known terminal equipment sampleMake Neural network classifier is trained, wherein m=1,2 for input using the corresponding device numbering m of each sample as output, 3 ..., M, n=1,2,3 ..., N.

5. the data packet legitimacy confirmation method according to claim 3 based on edge calculations and Discrete Stochastic convolution, It is characterized in that: in the step S10, by the channel matrices estimation set S of terminal device to be verified_m' in any sample be entered as In ripe neural network classifier, classification results are obtained, whether legal judge terminal device to be verified accordingly, to realize that this sets Standby data packet legitimacy confirmation.

Technical field

The present invention relates to the confirmations of the data packet legitimacy of communication system, more particularly to based on edge calculations and Discrete Stochastic The communication system data packet legitimacy confirmation method of convolution.

Background technique

The legitimacy confirmation of data packet access is to guarantee a kind of important method of edge calculations data packet transmission safety.Physics The legal confirmation method of data packet of layer channel characteristics is to judge to send by comparing the channel information similitude between successive frame Square identity information has advantage rapidly and efficiently, is very suitable for micro terminal equipment resource-constrained in edge calculations.Tradition The legal confirmation methods of data packet of channel characteristics determined that recognition accuracy is lower using threshold value, and it is unstable.Engineering It practises and deep learning is by the training to great amount of samples, obtain classifier, can effectively promote recognition accuracy.With how defeated Enter-multi output (MIMO) technology edge calculations transmission in popularize so that one-dimensional channel estimation vector has become receiver-channel The two-dimensional matrix sample (referred to as channel matrix) of estimated value.

Certain technologies suitable for image procossing can further promote knowledge if convolutional neural networks are applied to this Other accuracy rate.But receiver quantity is much smaller than the quantity of element in channel estimation vector, and receiver has certain independence (i.e. neighboring receivers and do not have higher correlation) obtains preferably so that convolutional neural networks not can be used directly in this Recognition result.

Summary of the invention

It is an object of the invention to overcome the deficiencies of the prior art and provide one kind to be based on edge calculations and Discrete Stochastic convolution Data packet legitimacy confirmation method, solve convolutional neural networks recognition effect be dissatisfactory in mimo channel matrix and ask Topic.

The purpose of the present invention is achieved through the following technical solutions: the number based on edge calculations and Discrete Stochastic convolution According to packet legitimacy confirmation method, comprising the following steps:

S1. unified original pilot signal X (p) is prestored in Edge Server and multiple known terminal equipment；

S2. known terminal equipment is inserted into known original pilot signal according to setting rule in the data packet for sending signal Afterwards, Edge Server is sent to using single antenna or multiple antennas；

S3. Edge Server receives the data packet that known terminal equipment is sent using multiple antennas receiving array, and to each The signal that antenna receives carries out pilot tone separation, the pilot signal matrix received:

Y (p)=[Y₁(p),Y₂(p),…,Y_K(p)]；

S4. Edge Server calculates the estimated value of channel matrix

Wherein, X (p)^-1Represent the inverse matrix of X (p)；

S5. it for any known terminal equipment, repeats step S2~S4 and carries out multiple data sampling and channel matrices estimation, Obtain the set of the estimated value of multiple channel matrixes:

Wherein,Represent the channel matrix obtained to m-th of known terminal equipment n-th sampled- data estimation；N=1, 2,3 ..., N, N indicate the number that data acquisition and channel matrices estimation are carried out to terminal device；

S6. for different known terminal equipment, step S2~S5 is repeated, the channel square of each known terminal equipment is obtained The set S of battle array estimation₁,S₂,…,S_M, wherein S_mRepresent the set of the channel matrices estimation of m-th of equipment, m=1,2,3 ..., M；

S7. according to the channel matrices estimation set S of each known terminal equipment₁,S₂,…,S_M, construct training set

S8. establish convolution kernel and convolution kernel movement rule: convolution kernel is continuous translation in longitudinal movement, laterally for from Dissipate randomized jitter:

S801. establishing convolution kernel is M ' × N ' matrix, expression are as follows:

CK=[L₁,L₂,…,L_N′]

Wherein L_n′The column vector for being M ' for dimension, M ' < M, N ' < N；

S802. convolution kernel longitudinal movement rule is identical as traditional convolutional neural networks, is continuous translation, and step-length is set as 1；

S803. convolution kernel transverse shifting rule is Discrete Stochastic bounce, specific as follows:

Convolution kernel initial position is CK₀=[L₁,L₂,…,L_N′], it is CK after the translation that traditional step-length is 1₁'=[L₂, L₃,…,L_N′+1], it will be appreciated that delete first row, and add last column right-hand column；Discrete Stochastic bounce rule are as follows: delete the One column, and the random column in last addition in addition to previous step convolution kernel includes column, it may be assumed that

S9. according to the convolution kernel and convolution kernel movement rule established in S8, neural network is constructed using convolutional neural networks Classifier, and classifier is trained using the training set in step S7, obtain mature neural network classifier；

S10. to terminal device to be verified, step S2~S5 is repeated, terminal device channel matrices estimation to be verified is obtained Set S_m′；Classified using mature neural network classifier to the channel matrix of terminal device to be verified, judges that it is It is no legal, to realize the data packet legitimacy confirmation of the equipment.

Preferably, the pilot signal X (p) is one-dimensional column matrix.

Wherein, the step S7 includes:

For the channel matrix set of each known terminal equipment, it is feedback with corresponding terminal device number, is instructed Practice setWherein, the corresponding training set of the channel matrix set of m-th of known terminal equipmentAre as follows:

Wherein, in the step S9, with each of the channel matrix set of each known terminal equipment sampleNeural network classifier is trained, wherein m as input using the corresponding device numbering m of each sample as output =1,2,3 ..., M, n=1,2,3 ..., N.

Wherein, in the step S10, by the channel matrices estimation set S of terminal device to be verified_m' in any sample it is defeated Enter in mature neural network classifier, obtains classification results, whether legal judge terminal device to be verified accordingly, to realize The data packet legitimacy of the equipment confirms.

The beneficial effects of the present invention are: the present invention passes through the convolution kernel of discrete convolutional neural networks, can randomly choose Discrete receiving antenna allows convolutional neural networks to obtain enough sample characteristics in receiving antenna dimension, improves The classifier of convolutional neural networks building recognition effect in mimo channel matrix, to improve the accuracy rate of identification.

Detailed description of the invention

Fig. 1 is flow chart of the method for the present invention；

Fig. 2 is that 8 hairs 8 receive mimo channel matrix waveform diagram in embodiment；

Fig. 3 is convolution kernel Discrete Stochastic movement rule schematic diagram.

Specific embodiment

Technical solution of the present invention is described in further detail with reference to the accompanying drawing, but protection scope of the present invention is not limited to It is as described below.

As shown in Figure 1, the data packet legitimacy confirmation method based on edge calculations and Discrete Stochastic convolution, including following step It is rapid:

S1. Edge Server and each terminal select unified pilot signal X (p) according to the rules, and wherein X (p) is one-dimensional column Matrix can use ZC sequence in embodiments herein；

S2. after terminal device is inserted into known original pilot signal according to setting rule in the data packet for sending signal, It is sent to Edge Server using single antenna or multiple antennas, in embodiments herein, by taking transmitting antenna number is 8 as an example；

S3. the data packet that Edge Server uses multiple antennas receiving array receiving terminal apparatus to send, and to each antenna The signal received carries out pilot tone separation, the pilot signal matrix received.

Y (p)=[Y₁(p),Y₂(p),…,Y₈(p)]

In embodiments herein, receiving antenna number is 8, as shown in Fig. 2, receiving mimo channel matrix waveform for 8 hairs 8 Figure.

S4. Edge Server calculates the estimated value of channel matrix

S5. it repeats S2~S4 more times, multiple data acquisition and channel matrices estimation is carried out to same terminal device, obtained more The set of the estimated value of a channel matrix.In embodiments herein, each equipment hits is 1000, then to first Know the set of the channel matrices estimation value of terminal device are as follows:

Represent the channel matrix that first equipment n-th sampled- data estimation obtains, n=1,2 ..., 1000；

S6. for different known terminal equipment, step S2~S5 is repeated, the channel square of each known terminal equipment is obtained The set of battle array estimation.In embodiments herein, number of devices 3 then obtains three until the channel matrix of terminal device is estimated Evaluation S₁,S₂,S₃, in which:

In formula, m=1,2,3；

S7. to the channel matrix set of each terminal device, it is feedback with corresponding terminal device number, obtains training set It closes

According to neural metwork training demand, m can be ten's digit, binary digit or 01 sequence.The application's In embodiment, m is corresponding 01 sequence of its device numbering, it may be assumed that

S8. establish convolution kernel and convolution kernel movement rule: convolution kernel is continuous translation in longitudinal movement, laterally for from Dissipate randomized jitter:

Specifically includes the following steps:

S801. establishing convolution kernel is M ' × N ' matrix, and in embodiments herein, convolution kernel is 40 × 3 matrixes, can table It reaches are as follows:

CK=[L₁,L₂,L₃]

Wherein L_n′The column vector for being 40 for dimension, as shown in Figure 3；

S802. convolution kernel longitudinal movement rule is identical as traditional convolutional neural networks, is continuous translation, and step-length is set as 1；

S803. convolution kernel transverse shifting rule is Discrete Stochastic bounce, as shown in figure 3, specific as follows:

Convolution kernel initial position is CK₀=[L₁,L₂,L₃], Discrete Stochastic bounce rule are as follows: delete first row, and last Add the random column in addition to previous step convolution kernel includes column, it may be assumed that

S9. with channel matrixFor training sample, number m is feedback, the convolution kernel and convolution kernel constructed according to S8 Movement rule is trained convolutional neural networks, obtains mature neural network classifier；

S10. to terminal device to be verified, step S2~S5 is repeated, terminal device channel matrices estimation to be verified is obtained Set S_m′；Classified using channel matrix of the trained neural network classifier to equipment to be verified, is determined belonging to it Equipment: by the channel matrices estimation set S of terminal device to be verified_m' in the mature neural network classifier of any sample input In, classification results are obtained, whether legal judge terminal device to be verified accordingly, to realize that the data packet legitimacy of the equipment is true Recognize, it is to be verified if the classification results of neural network classifier output are known device number in embodiments herein Terminal device is known device, and the data packet which sends is legal, and the data packet sent to the equipment receives；Otherwise, Equipment to be verified is illegality equipment, and the data packet which sends is illegal, abandons the data packet of equipment transmission.

The present invention passes through the convolution kernel of discrete convolutional neural networks, can randomly choose discrete receiving antenna, so that Convolutional neural networks can obtain enough sample characteristics in receiving antenna dimension, improve point of convolutional neural networks building Class device recognition effect in mimo channel matrix, to improve the accuracy rate of identification.

The above is a preferred embodiment of the present invention, it should be understood that the present invention is not limited to shape described herein Formula should not be viewed as excluding other embodiments, and can be used for other combinations, modification and environment, and can be in this paper institute It states in contemplated scope, modifications can be made through the above teachings or related fields of technology or knowledge.And what those skilled in the art were carried out Modifications and changes do not depart from the spirit and scope of the present invention, then all should be within the scope of protection of the appended claims of the present invention.