CN113395096B - Physical layer secure transmission method based on deep learning in FDD system - Google Patents

Physical layer secure transmission method based on deep learning in FDD system Download PDF

Info

Publication number
CN113395096B
CN113395096B CN202110702059.5A CN202110702059A CN113395096B CN 113395096 B CN113395096 B CN 113395096B CN 202110702059 A CN202110702059 A CN 202110702059A CN 113395096 B CN113395096 B CN 113395096B
Authority
CN
China
Prior art keywords
information
channel
signal
matrix
sender
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110702059.5A
Other languages
Chinese (zh)
Other versions
CN113395096A (en
Inventor
江峦
廖学文
尤蓉蓉
陈伟
何畅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hunan Guotian Electronic Technology Co ltd
Original Assignee
Hunan Guotian Electronic Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hunan Guotian Electronic Technology Co ltd filed Critical Hunan Guotian Electronic Technology Co ltd
Priority to CN202110702059.5A priority Critical patent/CN113395096B/en
Publication of CN113395096A publication Critical patent/CN113395096A/en
Application granted granted Critical
Publication of CN113395096B publication Critical patent/CN113395096B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
    • H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
    • H04B7/0413—MIMO systems
    • H04B7/0456—Selection of precoding matrices or codebooks, e.g. using matrices antenna weighting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
    • H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
    • H04B7/08—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the receiving station
    • H04B7/0837—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the receiving station using pre-detection combining
    • H04B7/0842—Weighted combining
    • H04B7/0848—Joint weighting
    • H04B7/0854—Joint weighting using error minimizing algorithms, e.g. minimum mean squared error [MMSE], "cross-correlation" or matrix inversion

Abstract

The invention discloses a physical layer safe transmission method based on deep learning in an FDD system, which comprises the following steps: designing a signal transmission model and an optimization problem; the information sender inputs the sending information into the coder and the quantizer, compresses and normalizes the estimated channel (-1,1), then carries out quantization coding, and sends out the compressed channel matrix coding after quantization; the information receiver inputs the received information into a decoder, and the decoder designs a corresponding pre-coding matrix according to the compressed channel information fed back by the information transmitter to ensure the safe transmission of the information; the information receiving side transmits a mixed signal of the target signal and the artificial noise to the information transmitting side. The invention has lower computational complexity, can effectively resist the eavesdropping of eavesdropping nodes, and the more the number of transmitting antennas is, the smaller the estimation error of the channel is, the more the number of feedback bits is, the better the performance of the system is; for SNR and estimation error
Figure 100004_DEST_PATH_IMAGE001
Has good robustness.

Description

Physical layer secure transmission method based on deep learning in FDD system
Technical Field
The invention belongs to the technical field of wireless communication, and particularly relates to a deep learning-based physical layer secure transmission method in an FDD system.
Background
The principle of Wireless Communication technology (Wireless Communication) is to utilize Wireless transceiver to realize the interconversion between information and electromagnetic signals and to realize the transmission process within a certain distance and range. The wide application of the wireless communication technology greatly improves the communication capability of human society and meets the communication requirements of various fields.
However, as the modern technology advances and demands increase, the application scale of the wireless communication device increases day by day, and the communication process faces many potential threats. Due to the broadcast nature of wireless transmissions, wireless transmissions are at risk of being eavesdropped by unauthorized eavesdropping nodes. In order to deal with the risk, a series of security technologies mainly based on key management appear, and a mechanism for ensuring the security is established on the basis of a computational cryptography method and mainly depends on the design of upper layer protocols in a computer to ensure the security of information. However, as the computing power of computers is continuously increasing, especially quantum computers are appearing, and the information transmission scenarios are more and more diversified, the traditional key mechanism is challenged more and more severely. The proposed physical layer security technology provides a new direction for solving the security problem of wireless communication, and unlike the traditional security technology, physical layer security is a technology for realizing wireless information transmission security by using the characteristics of communication equipment and channels, and more students in recent years carry out deep research on physical layer security communication from the information theory perspective.
For the research of the physical layer security, Goel and Negi of the university of Chiilong in the card introduce the concept of Artificial Noise (AN), and the artificial noise interferes AN eavesdropper by adding AN artificially generated noise signal into the null space of the channel of a legal receiver. A number of documents have studied the design of artificial noise schemes in various scenarios, and the royal coment of the university of west ann has proposed a framework for artificial noise assisted secure transmission in the mimo eavesdropping channel (mimo me) of Frequency Division Duplex (FDD) systems. And focuses on a practical scenario where only channel profile information (CDI) of an eavesdropper is available, whereas imperfect Channel State Information (CSI) of a legitimate receiver is obtained through training and analog feedback. By explicitly considering signaling overhead and training power overhead caused by channel estimation and feedback, the achievable effective traversal security rate is defined, and a joint power allocation and training overhead optimization problem is researched to realize the maximization of the effective traversal security rate. However, deriving a closed form expression for the optimization problem is a very difficult problem and mathematically problematic. Therefore, in the conventional method, power allocation or other parameters are often optimized by adopting a fixed precoding matrix design, so that not only the optimal precoding matrix conforming to the scene cannot be obtained, but also a large computational complexity exists. Due to the above problems, a new safe precoding scheme, namely a precoding design scheme based on deep learning, is proposed by the Sangseok Yun of the university of queen and the Jae-Mo Kang of the university of qing north, a convolutional neural network architecture is adopted, and under the condition that spatial channel correlation exists, a safe precoder is designed according to estimated (imperfect) channel information, and a two-step training program is also proposed to ensure that the network is trained to converge quickly and stably.
Disclosure of Invention
The invention provides a physical layer security transmission mode based on deep learning in a frequency division duplex system, which considers a transmitting node (information transmitting party) Alice, a legal receiving node (information receiving party) Bob and an eavesdropping node (eavesdropper) Eve. Since the system operates in a frequency division duplex system, the legitimate channels from Alice to Bob do not have reciprocity, i.e., the uplink (Bob to Alice) and downlink (Alice to Bob) use two separate frequencies and therefore have different channel characteristics. Because Alice cannot directly obtain the downlink channel information, Bob feeds back the downlink channel information to Alice, and then Alice designs a precoding matrix according to the fed-back channel information to ensure that the purpose of safe transmission is achieved.
To achieve the above object, the present invention uses a neural network to implement the process of feedback and precoding matrix design. Consider that Alice has
Figure DEST_PATH_IMAGE001
The root transmitting antenna, Bob and Eve are all single receiving antennas, wherein Alice goes to Bob's matrix for legal channel
Figure DEST_PATH_IMAGE002
Is expressed with a matrix size of
Figure DEST_PATH_IMAGE003
And each element thereof is independently distributed and obeys a zero mean variance of
Figure DEST_PATH_IMAGE004
Complex gaussian distribution of (a); and Alice to Eve eavesdropping channel matrix
Figure DEST_PATH_IMAGE005
Indicating that the matrix size is also
Figure 703350DEST_PATH_IMAGE003
And each element thereof is independently distributed and obeys a zero mean variance of
Figure DEST_PATH_IMAGE006
Complex gaussian distribution. At the same time, consider a more realistic situation, Bob is estimating
Figure DEST_PATH_IMAGE007
There is an estimation error, and Bob actually estimates the channel as
Figure DEST_PATH_IMAGE008
And satisfies the equation
Figure DEST_PATH_IMAGE009
Wherein
Figure DEST_PATH_IMAGE010
The error channel matrix generated when estimating the channel for Bob,
Figure 731743DEST_PATH_IMAGE008
and
Figure 714742DEST_PATH_IMAGE010
are all big and small
Figure DEST_PATH_IMAGE011
Assuming the estimation error is expressed as
Figure DEST_PATH_IMAGE012
，
Figure 93640DEST_PATH_IMAGE008
Can be approximately regarded as distribution-independent and obeys zero mean variance of
Figure DEST_PATH_IMAGE013
A complex Gaussian distribution of
Figure 704750DEST_PATH_IMAGE010
Can be regarded as distribution-independent and obeys zero mean variance of
Figure DEST_PATH_IMAGE014
Complex gaussian distribution.
The invention discloses a physical layer secure transmission method based on deep learning in an FDD system, which comprises the following steps:
designing a signal transmission model and an optimization problem according to the worst condition in wireless communication;
the information sending party inputs the sending information into the coder and the quantizer, and estimates the channel
Figure DEST_PATH_IMAGE015
Compression and normalization
Figure DEST_PATH_IMAGE016
Then carrying out quantization coding, and sending out the compressed channel matrix code after quantization;
the information receiver inputs the received information into the decoder, which designs the corresponding pre-coding matrix according to the compressed channel information fed back by the information transmitter
Figure DEST_PATH_IMAGE017
And
Figure DEST_PATH_IMAGE018
to ensure the secure transmission of information, wherein
Figure 507971DEST_PATH_IMAGE017
Is a signal sent from an information receiver to an information sender
Figure DEST_PATH_IMAGE019
The precoding matrix of (a) is determined,
Figure 430796DEST_PATH_IMAGE018
is an artificial noise signal added to a transmission signal by a signal receiving side
Figure DEST_PATH_IMAGE020
The precoding matrix of (a);
the information receiving side transmits a mixed signal of the target signal and the artificial noise to the information transmitting side.
Further, the optimization problem is modeled as follows:
assuming that the eavesdropper accurately estimates the channel
Figure DEST_PATH_IMAGE021
And precoding matrix, while signal receiver and signal transmitter only know the channel
Figure DEST_PATH_IMAGE022
The capacity expression of signal transmission is obtained as follows:
Figure DEST_PATH_IMAGE023
Figure DEST_PATH_IMAGE024
whereinR B Is the signal transmission capacity of the sender of the information,R E in order for the signal transmission capacity of the eavesdropper,
Figure DEST_PATH_IMAGE025
is a zero mean variance at the sender of the information,
Figure DEST_PATH_IMAGE026
is a zero mean variance at the eavesdropper,
Figure DEST_PATH_IMAGE027
and
Figure DEST_PATH_IMAGE028
respectively representing signals transmitted by information transmitters
Figure DEST_PATH_IMAGE029
And artificial noise signal
Figure DEST_PATH_IMAGE030
The digital pre-coding matrix of (2),
Figure DEST_PATH_IMAGE031
representing the analog precoding matrix produced by the phase shifter,
Figure DEST_PATH_IMAGE032
a matrix for a legitimate channel from an information receiver to an information sender,
Figure DEST_PATH_IMAGE033
for the estimated channel obtained by the sender of the information,
Figure DEST_PATH_IMAGE034
an error channel matrix generated when estimating a channel for an information transmitting side,
Figure DEST_PATH_IMAGE035
a matrix for an eavesdropping channel from an information receiver to an eavesdropper;
the privacy rate is calculated as follows:
Figure DEST_PATH_IMAGE036
the precoding matrix is expected to be designed to maximize the privacy rate, and the optimization problem can be expressed as follows:
Figure DEST_PATH_IMAGE037
Figure DEST_PATH_IMAGE038
。
further, the estimated channel of the information sender is used
Figure 502133DEST_PATH_IMAGE015
The real part and the imaginary part are separated and input into the encoder, the encoder is composed of three fully-connected layers, the activation function of the first two fully-connected layers is a sigmoid function, the activation function of the third fully-connected layer is a tanh function to ensure that the output is limited in the range of (-1,1), and the number of nodes of the third fully-connected layer is
Figure DEST_PATH_IMAGE039
To complete the estimation of the channel
Figure 702170DEST_PATH_IMAGE015
Compression of (2), wherein NTFor the number of elements per row of the precoding matrix,
Figure DEST_PATH_IMAGE040
is a compression factor.
Further, the quantizer performs quantization using B-bit, and the quantization process is as follows:
Figure DEST_PATH_IMAGE041
where B is the number of quantization bits,
Figure DEST_PATH_IMAGE042
the function of the rounding is represented by,
Figure DEST_PATH_IMAGE043
representing the elements contained in the compressed and normalized channel.
Further, the information receiving side receives the quantized signal fed back, and decodes the compressed and quantized feedback information through a decoder composed of three fully connected layers, and the decoder directly outputs the precoding matrix and the power distribution coefficient.
Further, it is assumed that the transmission antenna of the information receiving side is equipped with
Figure DEST_PATH_IMAGE044
A radio frequency chain, satisfy
Figure DEST_PATH_IMAGE045
，
Figure 272698DEST_PATH_IMAGE017
And
Figure 357853DEST_PATH_IMAGE018
can be replaced by the following equation:
Figure DEST_PATH_IMAGE046
Figure DEST_PATH_IMAGE047
wherein the content of the first and second substances,
Figure 953919DEST_PATH_IMAGE027
and
Figure 805201DEST_PATH_IMAGE028
respectively representing signals transmitted by information transmitters
Figure 438307DEST_PATH_IMAGE029
And artificial noise signal
Figure 7829DEST_PATH_IMAGE030
Number ofWord precoding matrix, and the matrix size are all
Figure DEST_PATH_IMAGE048
，
Figure 938745DEST_PATH_IMAGE031
Representing an analog precoding matrix produced by phase shifters, the matrix size of which is
Figure DEST_PATH_IMAGE049
，NTThe number of elements per row of the precoding matrix.
Further, the received signals of the information sender and the eavesdropper can be expressed as:
Figure DEST_PATH_IMAGE050
Figure DEST_PATH_IMAGE051
Figure DEST_PATH_IMAGE052
wherein the content of the first and second substances,y B a received signal indicating a sender of the information,y E a received signal representing an eavesdropper,
Figure DEST_PATH_IMAGE053
representing receiver noise at the sender of the information and subject to zero mean variance of
Figure DEST_PATH_IMAGE054
The complex gaussian distribution of (a) is,
Figure DEST_PATH_IMAGE055
representing receiver noise of an eavesdropper and obeying a zero mean variance of
Figure 359712DEST_PATH_IMAGE026
The complex gaussian distribution of (a) is,
Figure 22774DEST_PATH_IMAGE029
is an energy normalized signal and satisfies
Figure DEST_PATH_IMAGE056
，
Figure 141909DEST_PATH_IMAGE030
Obeying a zero mean variance of
Figure DEST_PATH_IMAGE057
The complex gaussian distribution of (a) is,
Figure 17461DEST_PATH_IMAGE032
a matrix for a legitimate channel from an information receiver to an information sender,
Figure 580685DEST_PATH_IMAGE033
for the estimated channel obtained by the sender of the information,
Figure 680228DEST_PATH_IMAGE034
an error channel matrix generated when estimating a channel for an information transmitting side,
Figure 834129DEST_PATH_IMAGE035
the matrix is used for an eavesdropping channel from an information receiver to an eavesdropper.
Further, the channel will be estimated at the time of training
Figure 44530DEST_PATH_IMAGE033
Is regarded as a true channel, and since the information sender and the information receiver only know the matrix for the eavesdropping channel from the information receiver to the eavesdropper
Figure 990490DEST_PATH_IMAGE035
The statistical information of (1), a channel with the same distribution is generated during training
Figure DEST_PATH_IMAGE058
As an eavesdropping channel, the loss function is therefore expressed as follows:
Figure DEST_PATH_IMAGE059
wherein the content of the first and second substances,
Figure 792092DEST_PATH_IMAGE054
and
Figure 557923DEST_PATH_IMAGE026
is zero mean variance;
and after the loss function is trained, using the secret keeping rate as a performance index of the test set.
Further, the normalization process is as follows:
for the output
Figure DEST_PATH_IMAGE060
Adding a lambda layer to satisfy
Figure DEST_PATH_IMAGE061
As follows:
Figure DEST_PATH_IMAGE062
output with two nodes
Figure DEST_PATH_IMAGE063
And
Figure DEST_PATH_IMAGE064
and normalizing to obtain the power distribution coefficient
Figure DEST_PATH_IMAGE065
The power distribution coefficient
Figure 451578DEST_PATH_IMAGE065
The normalization process is as follows:
Figure DEST_PATH_IMAGE066
to satisfy
Figure DEST_PATH_IMAGE067
To, for
Figure DEST_PATH_IMAGE068
Performing normalization process, supposing
Figure DEST_PATH_IMAGE069
To represent
Figure DEST_PATH_IMAGE070
，
Figure 32470DEST_PATH_IMAGE070
The normalization process is as follows:
Figure DEST_PATH_IMAGE071
suppose to use
Figure DEST_PATH_IMAGE072
To represent
Figure DEST_PATH_IMAGE073
To satisfy
Figure DEST_PATH_IMAGE074
，W BB The normalization process is as follows:
Figure DEST_PATH_IMAGE075
the physical layer safe transmission mode based on deep learning in the FDD system disclosed by the invention has the following advantages:
1) compared with a physical layer safety scheme under the traditional FDD system, the method has lower computational complexity by using a neural network design safety scheme.
2) Simulation proves that the invention can effectively resist eavesdropping of eavesdropping nodes, and the more the number of transmitting antennas is, the smaller the estimation error of the channel is, the more the number of feedback bits is, the better the performance of the system is.
3) Simulation results prove that the neural network designed by the invention has good SNR and estimation error
Figure 460433DEST_PATH_IMAGE012
Has good robustness.
Drawings
FIG. 1 is a diagram of a transmission process in an FDD-MISO system;
FIG. 2 is a schematic diagram of a neural network structure of a physical layer secure transmission scheme according to the present invention;
FIG. 3 shows the present invention
Figure DEST_PATH_IMAGE076
，
Figure DEST_PATH_IMAGE077
And
Figure DEST_PATH_IMAGE078
the effect of signal-to-noise ratio (SNR) on the privacy rate;
FIG. 4 shows the present invention
Figure 713560DEST_PATH_IMAGE076
，
Figure 531343DEST_PATH_IMAGE077
And
Figure DEST_PATH_IMAGE079
estimate error in the case of
Figure 121069DEST_PATH_IMAGE012
The effect on the privacy rate;
FIG. 5 shows the present invention
Figure DEST_PATH_IMAGE080
，
Figure 326791DEST_PATH_IMAGE077
，
Figure DEST_PATH_IMAGE081
And
Figure 270477DEST_PATH_IMAGE078
number of feedback bits in case of
Figure DEST_PATH_IMAGE082
The effect on the privacy rate.
Detailed Description
The invention is further described with reference to the accompanying drawings, but the invention is not limited in any way, and any alterations or substitutions based on the teaching of the invention are within the scope of the invention.
The invention is described in further detail below with reference to the accompanying drawings:
the transmission process of the present invention is shown in FIG. 1, considering that Alice has
Figure 954268DEST_PATH_IMAGE001
The root transmitting antenna, Bob and Eve are all single receiving antennas, wherein the matrix for the legal channels from Alice to Bob
Figure 339113DEST_PATH_IMAGE002
Is expressed with a matrix size of
Figure 122261DEST_PATH_IMAGE011
And each element thereof is independently distributed and obeys a zero mean variance of
Figure 349980DEST_PATH_IMAGE004
Complex gaussian distribution of (a); and Alice to Eve eavesdropping channel matrix
Figure 119353DEST_PATH_IMAGE005
Indicating that the matrix size is also
Figure 17426DEST_PATH_IMAGE003
And each element thereof is independently distributed and obeys a zero mean variance of
Figure 705896DEST_PATH_IMAGE006
Complex gaussian distribution. At the same time, consider a more realistic situation, Bob is estimating
Figure 296278DEST_PATH_IMAGE002
There is an estimation error, and Bob actually estimates the channel as
Figure 259555DEST_PATH_IMAGE008
And satisfies the equation
Figure DEST_PATH_IMAGE083
Wherein
Figure 681309DEST_PATH_IMAGE010
The error channel matrix generated when estimating the channel for Bob,
Figure DEST_PATH_IMAGE084
and
Figure DEST_PATH_IMAGE085
are all big and small
Figure 462052DEST_PATH_IMAGE003
Assuming the estimation error is expressed as
Figure 274150DEST_PATH_IMAGE014
，
Figure 41118DEST_PATH_IMAGE008
Can be approximately regarded as distribution-independent and obeys zero mean variance of
Figure DEST_PATH_IMAGE086
A complex Gaussian distribution of
Figure 580028DEST_PATH_IMAGE010
Can be regarded as distribution-independent and obeys zero mean variance of
Figure 875880DEST_PATH_IMAGE012
Complex gaussian distribution. The transmission process of the signal is as follows:
1) bob will estimate the channel
Figure 565487DEST_PATH_IMAGE008
Figure 480354DEST_PATH_IMAGE084
Compressed and normalized to
Figure DEST_PATH_IMAGE087
Then the quantization coding is carried out on the data,Bthe expression for bit quantization is as follows:
Figure DEST_PATH_IMAGE088
(1)
wherein
Figure DEST_PATH_IMAGE089
The function of the rounding is represented by,
Figure DEST_PATH_IMAGE090
representing the elements contained in the compressed and normalized channel. In the formula (1), first
Figure 860388DEST_PATH_IMAGE090
Is extended to
Figure DEST_PATH_IMAGE091
And then rounded to the nearest integer. Then, in order to guarantee
Figure 795983DEST_PATH_IMAGE090
Is a quiltBQuantization of bits by
Figure DEST_PATH_IMAGE092
Instead of being rounded to
Figure DEST_PATH_IMAGE093
Of (2) is used. Finally by division by
Figure DEST_PATH_IMAGE094
Make the output
Figure DEST_PATH_IMAGE095
Has a value range of
Figure DEST_PATH_IMAGE096
. With this method, each element in the channel matrix can be processedB-bit quantization. The invention adopts 4-bit quantization, and Bob sends the compressed channel matrix code after 4-bit quantization to Alice.
2) Alice designs corresponding precoding matrix according to compressed channel information fed back by Bob
Figure DEST_PATH_IMAGE097
And
Figure DEST_PATH_IMAGE098
to ensure the secure transmission of information, wherein
Figure 21822DEST_PATH_IMAGE097
Is the signal sent by Alice to Bob
Figure DEST_PATH_IMAGE099
Of a size of
Figure DEST_PATH_IMAGE100
。
Figure DEST_PATH_IMAGE101
Is an artificial noise signal added by Alice in the transmitted signal
Figure DEST_PATH_IMAGE102
And the same size of the precoding matrix of
Figure 49034DEST_PATH_IMAGE100
. The present invention considers a hybrid precoding scheme, assuming Alice's transmit antenna is equipped with
Figure DEST_PATH_IMAGE103
A radio frequency chain, satisfy
Figure DEST_PATH_IMAGE104
，
Figure 158941DEST_PATH_IMAGE097
And
Figure DEST_PATH_IMAGE105
can be replaced by the following equation:
Figure DEST_PATH_IMAGE106
(2)
Figure DEST_PATH_IMAGE107
(3)
wherein the content of the first and second substances,
Figure DEST_PATH_IMAGE108
and
Figure DEST_PATH_IMAGE109
respectively represent
Figure DEST_PATH_IMAGE110
And
Figure DEST_PATH_IMAGE111
and the matrix size is all
Figure DEST_PATH_IMAGE112
。
Figure DEST_PATH_IMAGE113
Representing an analog precoding matrix produced by phase shifters, the matrix size of which is
Figure DEST_PATH_IMAGE114
。
3) Alice transmits a mixed signal of the target signal and the artificial noise to Bob, and Eve also receives the mixed signal transmitted by Alice due to the broadcasting characteristics of wireless communication. Thus, the received signals of Bob and Eve may be expressed as:
Figure DEST_PATH_IMAGE115
Figure DEST_PATH_IMAGE116
Figure DEST_PATH_IMAGE117
(4)
wherein the content of the first and second substances,
Figure DEST_PATH_IMAGE118
represents the receiver noise at Bob and obeys a zero mean variance of
Figure DEST_PATH_IMAGE119
Complex gaussian distribution.
Figure DEST_PATH_IMAGE120
Represents receiver noise at Eve and obeys a zero mean variance of
Figure DEST_PATH_IMAGE121
Complex gaussian distribution.
Figure 505519DEST_PATH_IMAGE110
Is an energy normalized signal and satisfies
Figure DEST_PATH_IMAGE122
。
Figure 497615DEST_PATH_IMAGE111
Obeying a zero mean variance of
Figure DEST_PATH_IMAGE123
Complex gaussian distribution. Suppose that Alice transmits a signal at a transmission power ofPThe power distribution coefficient to the target signal and the artificial noise is
Figure DEST_PATH_IMAGE124
Then there are
Figure DEST_PATH_IMAGE125
And
Figure DEST_PATH_IMAGE126
。
4) considering the worst condition in wireless communication, Eve can accurately estimate the channel
Figure 941234DEST_PATH_IMAGE005
And precoding matrices, while Alice and Bob only know
Figure DEST_PATH_IMAGE127
The statistical information of (1). According to the transmission process and the expression, capacity expressions of Alice-to-Bob and Eve signal transmission can be obtained respectively as follows:
Figure DEST_PATH_IMAGE128
(5)
Figure DEST_PATH_IMAGE129
Figure DEST_PATH_IMAGE130
(6)
the secret rate can then be obtained by the difference between:
Figure DEST_PATH_IMAGE131
(7)
Figure DEST_PATH_IMAGE132
as an index for measuring security performance of the present invention, it is desirable that the designed precoding matrix can be maximized to achieve the purpose of transmission security, so the optimization problem can be expressed as:
Figure DEST_PATH_IMAGE133
Figure DEST_PATH_IMAGE134
(8)
the present invention will next use neural networks to accomplish the above transmission process and maximize the privacy rate.
Fig. 2 shows the structure of the neural network of the present invention. It can be seen that the neural network is made up of three parts, an encoder part and a quantizer part at Bob, and a decoder part at Alice.
1) An encoder: bob will estimate
Figure 16900DEST_PATH_IMAGE008
The real part and the imaginary part are separated and input into the neural network, so that the input of the neural network is
Figure DEST_PATH_IMAGE135
. The first layer full connection layer FC1 of the encoder is set to be weighted
Figure DEST_PATH_IMAGE136
The offset number is set as
Figure DEST_PATH_IMAGE137
The activation function is a sigmoid function, so the output size of FC1 is
Figure DEST_PATH_IMAGE138
. FC2 has a weight of
Figure DEST_PATH_IMAGE139
Is offset fromThe number is set as
Figure 212739DEST_PATH_IMAGE138
The activation function is also a sigmoid function, with an output magnitude of
Figure 98656DEST_PATH_IMAGE137
. Unlike FC1 and FC2, FC3 completes compression and normalization of channel information, and thus the number of weights is set to be
Figure DEST_PATH_IMAGE140
The offset number is set as
Figure DEST_PATH_IMAGE141
So that the output size is compressed into
Figure 814808DEST_PATH_IMAGE141
(ii) a In addition to this, the activation function is set to tanh function so that all outputs of FC3 are limited to
Figure 977936DEST_PATH_IMAGE087
Within the range of (1).
2) A quantizer: when FC3 restricts the compressed matrix to
Figure 957393DEST_PATH_IMAGE087
The output of the FC3 can be quantized to the signal x using the formula (1). It is noted that the rounding function in the quantization operation
Figure DEST_PATH_IMAGE142
It is not trivial and therefore not possible to compute the gradient of the quantization module when propagating backwards. To solve this problem, we set the gradient of the quantization process constant to 1, so that the entire neural network can be trained.
3) A decoder: the bit number of the feedback received by Alice is
Figure DEST_PATH_IMAGE143
And as an output of the decoderAnd (6) adding. Therefore, the FC4 weight number is set to
Figure DEST_PATH_IMAGE144
Offset number is
Figure DEST_PATH_IMAGE145
The activation function is a sigmoid function, and the final matrix size of the output is
Figure 786065DEST_PATH_IMAGE145
. FC5 has a weight of
Figure DEST_PATH_IMAGE146
Offset number is
Figure 712433DEST_PATH_IMAGE138
The activation function is also a sigmoid function, and the output has a magnitude of
Figure 589122DEST_PATH_IMAGE138
. FC6 is used as an output layer without an activation function, and the weight number is equal to the output matrix size and the power distribution parameter
Figure DEST_PATH_IMAGE147
Offset number is
Figure DEST_PATH_IMAGE148
The output is also of the same size
Figure DEST_PATH_IMAGE149
。
4) Normalization: the output of the neural network needs to be normalized to meet the energy constraints in the optimization problem (8). For the output
Figure DEST_PATH_IMAGE150
Adding a lambda layer to satisfy
Figure DEST_PATH_IMAGE151
As follows:
Figure DEST_PATH_IMAGE152
(9)
then, since the output layer has no activation function, the output range cannot be within by one node
Figure DEST_PATH_IMAGE153
Power distribution coefficient of
Figure DEST_PATH_IMAGE154
Thus consider outputting with two nodes
Figure DEST_PATH_IMAGE155
And
Figure DEST_PATH_IMAGE156
and normalized to obtain
Figure 579294DEST_PATH_IMAGE124
And the normalization process comprises the following steps:
Figure DEST_PATH_IMAGE157
(10)
then, to satisfy
Figure 502120DEST_PATH_IMAGE125
Need to be aligned with
Figure DEST_PATH_IMAGE158
Performing normalization process, supposing
Figure DEST_PATH_IMAGE159
To represent
Figure DEST_PATH_IMAGE160
The normalization process is as follows:
Figure DEST_PATH_IMAGE161
(11)
in the same wayAssume use of
Figure DEST_PATH_IMAGE162
To represent
Figure DEST_PATH_IMAGE163
To satisfy
Figure DEST_PATH_IMAGE164
The treatment process comprises the following steps:
Figure DEST_PATH_IMAGE165
(12)
5) loss function: the input of the neural network is
Figure 468064DEST_PATH_IMAGE008
The network is not in the training process
Figure 74626DEST_PATH_IMAGE085
And therefore cannot construct a loss function based on the true privacy rate of equation (7). Considering the actual situation, the training will be carried out
Figure 661465DEST_PATH_IMAGE084
Is viewed as a real channel and since Alice and Bob only know
Figure 743690DEST_PATH_IMAGE127
The statistical information of (1), a channel with the same distribution is generated during training
Figure DEST_PATH_IMAGE166
As an eavesdropping channel, the loss function can therefore be written as:
Figure DEST_PATH_IMAGE167
(13)
Figure 339757DEST_PATH_IMAGE054
and
Figure 188108DEST_PATH_IMAGE026
is zero mean variance. After training using the loss function of equation (13), the true privacy rate of equation (7) is used as the performance index of the test set.
For the convenience of the simulation, assume that
Figure DEST_PATH_IMAGE168
，
Figure DEST_PATH_IMAGE169
，
Figure DEST_PATH_IMAGE170
，
Figure DEST_PATH_IMAGE171
. At the time of training, the blocksize was set to 1000, the number of iterations was set to 30000, the learning rate was set to 0.05, and the Adam optimizer was used, resulting in the following simulation graph.
Case 1: during training, SNR is set to 10dB, and error is estimated
Figure 273745DEST_PATH_IMAGE012
Set to 0.01; estimate error at test time
Figure 577687DEST_PATH_IMAGE012
Again set to 0.01 and the number of feedback bits to
Figure DEST_PATH_IMAGE172
bits, the effect of SNR on the secret rate shown in fig. 3 is obtained. As can be seen from fig. 3, the SNR increases and the secret rate of the system increases, because the transmission capacity from Alice to Bob increases significantly due to the increase of Alice transmission power, but the transmission capacity from Alice to Eve increases insignificantly due to the presence of artificial noise. In addition, comparing the three curves in fig. 3, it can be seen that the transmitting antenna also has a great influence on the security performance of the system, and
Figure 915128DEST_PATH_IMAGE001
the security performance of the system is improved due to the fact that the number of the antennas is increased, the gain of a target signal part at the receiving end is increased, namely the received signal-to-interference-and-noise ratio (SINR) is increased, and therefore the transmission capacity from Alice to Bob is increased. Therefore, to improve security performance, adding transmit antennas is a viable option.
Case 2: in FIG. 4, the SNR in training is set to 10dB, and the estimation error in training is set to 10dB
Figure 886495DEST_PATH_IMAGE014
Is 0.01; SNR is 10dB and the number of feedback bits is
Figure 549557DEST_PATH_IMAGE172
bits. From the illustration in fig. 4, it can be seen that the presence of estimation errors has a significant impact on the privacy performance, and as the estimation errors increase, the privacy rate gradually decreases, due to the impact of both aspects of the estimation errors: (1) the estimation error is increased, so that the difference between the expression of the secret rate and the loss function used in training is increased, and the optimal weight and bias obtained by training and the optimal weight and bias needed for optimizing the secret rate are increased. (2) The increase in estimation error results in equation (15)
Figure DEST_PATH_IMAGE173
Increase with transmission capacity from Alice to Bob
Figure DEST_PATH_IMAGE174
And thus the privacy rate is reduced. Bob channel estimation accuracy is therefore critical to system privacy performance.
Case 3: the parameters of fig. 5 are set as: SNR during training is 10dB, estimation error during training
Figure 406042DEST_PATH_IMAGE014
0.01, an estimation error SNR at the time of test of 10dB, an estimation error at the time of test of
Figure 812753DEST_PATH_IMAGE014
Is 0.01. It can be seen that with the number of feedback bits
Figure 513993DEST_PATH_IMAGE143
The increase, the security performance is obviously improved, this is because: suppose the FC3 node number is composed of
Figure DEST_PATH_IMAGE175
Meaning that after 4-bit quantization, the quantizer outputs a value of
Figure DEST_PATH_IMAGE176
And (4) possibility. Therefore if the number of bits is fed back
Figure 144694DEST_PATH_IMAGE143
Increase means that
Figure 16704DEST_PATH_IMAGE175
The probability of the value output by the quantizer is increased, and the larger optimizing space enables the network to find the optimal precoding matrix and the power distribution coefficient more easily, so that the confidentiality rate of the system is improved.
Therefore, in conclusion, the deep learning-based physical layer secure transmission method in the FDD-MISO system can effectively resist eavesdropping.
The physical layer safe transmission mode based on deep learning in the FDD system disclosed by the invention has the following advantages:
1) compared with a physical layer safety scheme under the traditional FDD system, the method has lower computational complexity by using a neural network design safety scheme.
2) Simulation proves that the invention can effectively resist eavesdropping of eavesdropping nodes, and the more the number of transmitting antennas is, the smaller the estimation error of the channel is, the more the number of feedback bits is, the better the performance of the system is.
3) Simulation results prove that the neural network designed by the invention has good SNR and estimation error
Figure 227106DEST_PATH_IMAGE014
Has the advantages ofGood robustness.
The above embodiment is an embodiment of the present invention, but the embodiment of the present invention is not limited by the above embodiment, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be regarded as equivalent replacements within the protection scope of the present invention.

Claims (9)

1. A deep learning-based physical layer secure transmission method in an FDD system is characterized by comprising the following steps:
designing a signal transmission model and an optimization problem according to the worst condition in wireless communication;
the information sender inputs the sending information into the coder and the quantizer, wherein the estimated channel of the information sender
Figure FDA0003487387510000011
The real part and the imaginary part are separated and input into the coder to estimate the channel
Figure FDA0003487387510000012
After compression and normalization, carrying out quantization coding, and sending out the compressed channel matrix code after quantization;
an information receiving party inputs received information into a decoder, and the decoder designs corresponding pre-coding matrixes F and W according to compressed channel information fed back by an information sending party to ensure the safe transmission of the information, wherein F is a pre-coding matrix of a signal s sent by the information sending party from an information receiving party, and W is a pre-coding matrix of an artificial noise signal v added in the sent signal by the signal receiving party;
the information receiver sends a mixed signal of the target signal and the artificial noise to the information sender;
wherein the optimization problem is modeled as follows: assuming that an eavesdropper accurately estimates and obtains a channel g and a precoding matrix, a signal receiver and a signal transmitter only know the statistical information of the channel g to obtain a capacity expression of signal transmission, and expecting to design the precoding matrix to maximize the confidentiality rate and establish an optimization problem;
wherein the channel is estimated during training
Figure FDA0003487387510000013
The channel is regarded as a real channel, and because the information sender and the information receiver only know the statistical information of the matrix g for the eavesdropping channel from the information receiver to an eavesdropper, a channel g' with the same distribution is generated as the eavesdropping channel during training, and a loss function is obtained.
2. The method for deep learning-based physical layer secure transmission in an FDD system according to claim 1, wherein the capacity expression of the signal transmission is as follows:
Figure FDA0003487387510000014
Figure FDA0003487387510000015
wherein R isBIs the signal transmission capacity, R, of the information senderEIn order for the signal transmission capacity of the eavesdropper,
Figure FDA0003487387510000016
is a zero mean variance at the sender of the information,
Figure FDA0003487387510000021
is the variance of zero mean at the eavesdropper, FBBAnd WBBDigital precoding matrices, F, representing respectively a signal s and an artificial noise signal v transmitted by an information senderRFRepresenting the analog precoding matrix generated by the phase shifter, h is the matrix for the legitimate channel from the information receiver to the information sender,
Figure FDA0003487387510000022
estimated channel, h, obtained for the sender of the informationerrorG is a matrix for an eavesdropping channel from an information receiver to an eavesdropper;
the privacy rate is calculated as follows:
Rs＝RB-RE
the optimization problem is represented as follows:
Figure FDA0003487387510000023
s.t.|FRF(i，j)|2＝1，||FRFFBB||2＝θP，||FRFWBB||2＝(1-θ)P，
wherein, P is the transmission power of the information receiver in the process of sending the signal, and theta is the power distribution coefficient distributed to the target signal and the artificial noise.
3. The method for deep-learning-based physical layer secure transmission in an FDD system according to claim 1, wherein the encoder comprises three fully-connected layers, the activation function of the first two fully-connected layers is sigmoid function, the activation function of the third fully-connected layer is tanh function to ensure that its output is limited to the range of (-1,1), and the number of nodes of the third fully-connected layer is
Figure FDA0003487387510000024
To complete the estimation of the channel
Figure FDA0003487387510000025
Compression of (2), wherein NTGamma is the compression factor, the number of elements per row of the precoding matrix.
4. The method for deep learning based secure transmission of the physical layer in the FDD system as claimed in claim 1 wherein the quantizer uses B-bit quantization, the quantization process is as follows:
y＝round(2B-1×x)/2B-1
where B is the number of quantization bits, round (-) represents the rounding function, and x represents the elements contained in the compressed and normalized channel.
5. The method for deep learning-based physical layer secure transmission in an FDD system according to claim 1, characterized in that the information receiver receives the quantized signal from the feedback and decodes the compressed quantized feedback information by a decoder consisting of three fully connected layers, which directly outputs the precoding matrix and the power allocation coefficients.
6. The FDD system deep learning-based physical layer secure transmission method according to claim 2, wherein it is assumed that N is allocated to the transmitting antenna of the information receiverRFA radio frequency chain satisfying NRF＜NTF and W may be replaced by the following equation:
F＝FRFFBB
W＝FRFWBB
wherein, FBBAnd WBBDigital pre-coding matrixes respectively representing signals s and artificial noise signals v sent by an information sender, wherein the sizes of the matrixes are NRF×1，FRFRepresenting an analog precoding matrix produced by phase shifters, the matrix size of which is NT×NRF，NTThe number of elements per row of the precoding matrix.
7. The method for deep learning-based physical layer secure transmission in FDD system according to claim 6, wherein the received signals of the information sender and the eavesdropper can be expressed as:
Figure FDA0003487387510000031
yE＝gFRFFBBs+gFRFWBBv+nE
wherein, yBIndicating the received signal of the sender of the information, yEReceived signal representing an eavesdropper, nBRepresenting receiver noise at the sender of the information and subject to zero mean variance of
Figure FDA0003487387510000032
Complex gaussian distribution of (n)ERepresenting receiver noise of an eavesdropper and obeying a zero mean variance of
Figure FDA0003487387510000033
S is an energy normalized signal and satisfies
Figure FDA0003487387510000034
v obeys a complex Gaussian distribution with a zero mean variance of 1, h is a matrix for a legal channel from an information receiver to an information sender,
Figure FDA0003487387510000035
estimated channel, h, obtained for the sender of the informationerrorG is a matrix for an eavesdropping channel from an information receiver to an eavesdropper.
8. The method for deep learning-based physical layer secure transmission in an FDD system according to claim 7, characterized in that the loss function is expressed as follows:
Figure FDA0003487387510000041
and after the loss function is trained, using the secret keeping rate as a performance index of the test set.
9. The method for deep learning-based physical layer secure transmission in an FDD system according to claim 8, characterized in that the normalization process is as follows:
for the output
Figure FDA0003487387510000042
Adding a lambda layer to satisfy | FRF(i，j)|21, as follows:
Figure FDA0003487387510000043
output theta with two nodes1And theta2And normalizing to obtain the power distribution coefficient theta, wherein the power distribution coefficient theta is normalized as follows:
Figure FDA0003487387510000044
to satisfy | | FRFFBB||2θ P, pair
Figure FDA0003487387510000045
Normalization was performed assuming FoTo represent
Figure FDA0003487387510000046
Figure FDA0003487387510000047
The normalization process is as follows:
Figure FDA0003487387510000048
suppose using WoTo represent
Figure FDA0003487387510000049
To satisfy | | FRFWBB||2＝(1-θ)P，WBBThe normalization process is as follows:
Figure FDA00034873875100000410
CN202110702059.5A 2021-06-24 2021-06-24 Physical layer secure transmission method based on deep learning in FDD system Active CN113395096B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110702059.5A CN113395096B (en) 2021-06-24 2021-06-24 Physical layer secure transmission method based on deep learning in FDD system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110702059.5A CN113395096B (en) 2021-06-24 2021-06-24 Physical layer secure transmission method based on deep learning in FDD system

Publications (2)

Publication Number Publication Date
CN113395096A CN113395096A (en) 2021-09-14
CN113395096B true CN113395096B (en) 2022-03-18

Family

ID=77623688

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110702059.5A Active CN113395096B (en) 2021-06-24 2021-06-24 Physical layer secure transmission method based on deep learning in FDD system

Country Status (1)

Country Link
CN (1) CN113395096B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113765556A (en) * 2021-09-27 2021-12-07 国家计算机网络与信息安全管理中心 Data transmission method and related equipment
CN113746628B (en) * 2021-09-30 2022-07-12 西安交通大学 Physical layer key generation method and system based on deep learning
CN114884616B (en) * 2022-05-06 2024-01-19 西安交通大学 Anti-eavesdrop uplink transmission method and device for satellite communication system based on artificial noise

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109194378A (en) * 2018-08-06 2019-01-11 重庆邮电大学 Safety of physical layer beam form-endowing method based on linear neural network
CN109246048A (en) * 2018-10-30 2019-01-18 广州海格通信集团股份有限公司 A kind of safety of physical layer communication means and system based on deep learning
CN110139272A (en) * 2019-05-21 2019-08-16 电子科技大学 A kind of secret signalling based on random sequence four-dimension ring array
CN111082933A (en) * 2019-12-27 2020-04-28 电子科技大学 Multi-user physical layer safety communication method capable of resisting any plurality of cooperation eavesdroppers

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10069592B1 (en) * 2015-10-27 2018-09-04 Arizona Board Of Regents On Behalf Of The University Of Arizona Systems and methods for securing wireless communications
EP3535625B1 (en) * 2016-12-07 2021-02-24 Arilou Information Security Technologies Ltd. System and method for using signal waveform analysis for detecting a change in a wired network

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109194378A (en) * 2018-08-06 2019-01-11 重庆邮电大学 Safety of physical layer beam form-endowing method based on linear neural network
CN109246048A (en) * 2018-10-30 2019-01-18 广州海格通信集团股份有限公司 A kind of safety of physical layer communication means and system based on deep learning
CN110139272A (en) * 2019-05-21 2019-08-16 电子科技大学 A kind of secret signalling based on random sequence four-dimension ring array
CN111082933A (en) * 2019-12-27 2020-04-28 电子科技大学 Multi-user physical layer safety communication method capable of resisting any plurality of cooperation eavesdroppers

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Secure Transmission in FDD MBM Systems Using Analog Feedback;Xiaomeng Zhou 等;《2020 IEEE/CIC International Conference on Communications in China (ICCC)》;20201109;全文 *
基于设备与信道特征的物理层安全方法;李古月 等;《密码学报》;20200415;第7卷(第2期);第224-248页 *
智能化物理层安全认证及传输技术研究;邱晓英;《中国博士学位论文全文数据库(信息科技辑)》;20190815;正文第3-4节 *

Also Published As

Publication number Publication date
CN113395096A (en) 2021-09-14

Similar Documents

Publication Publication Date Title
CN113395096B (en) Physical layer secure transmission method based on deep learning in FDD system
Sohrabi et al. Deep learning for distributed channel feedback and multiuser precoding in FDD massive MIMO
CN109194378B (en) Physical layer safety wave beam shaping method based on linear neural network
CN109150855B (en) Robust wireless communication safety transmission method for optimizing power resources
CN110880950A (en) Safe transmission method for artificial noise auxiliary vector disturbance precoding in MIMO system
Sindhwani et al. Performance analysis of particle swarm optimization and genetic algorithm in MIMO systems
Yu et al. Non-orthogonal wireless backhaul design for cell-free massive MIMO: An integrated computation and communication approach
CN110798282B (en) Artificial noise implementation method of software radio MISO system
CN112583458A (en) MIMO end-to-end transmission system based on deep learning and wireless transformation network
Wang et al. Multicast secrecy rate maximization for reconfigurable intelligent surface backscatter communication
Zhao et al. MSE-based transceiver designs for RIS-aided communications with hardware impairments
CN109714090B (en) Multi-antenna two-dimensional vector transmission method and system
CN116760439A (en) Robust wave beam forming method based on radar communication integrated system
CN114286336B (en) Multi-cell network safe transmission method based on artificial noise
CN113612571B (en) Multi-component safe transmission method based on generalized multi-fraction Fourier transform
CN114337902B (en) IRS-assisted millimeter wave multi-cell interference suppression method
CN114006639B (en) Distributed precoding method suitable for uplink multi-user MIMO system
CN112636799B (en) Optimal pseudo noise power configuration method in MIMO (multiple input multiple output) safety communication
Xu et al. Performance evaluation of autoencoder for coding and modulation in wireless communications
CN113364502B (en) Physical layer secure communication method between multi-antenna equipment
CN115208443B (en) Combined optimization method and device for large-scale MIMO system based on intelligent reflecting surface
CN117528535B (en) Bidirectional safe transmission method for resisting malicious interference by using IRS and artificial noise
Chen et al. Joint design of interference alignment and power splitting in SWIPT networks
CN114430550B (en) Wireless physical layer key generation method based on neural network model
Yang et al. Optimal Beamforming for Secure Transmit in Practical Wireless Networks

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
PE01 Entry into force of the registration of the contract for pledge of patent right
PE01 Entry into force of the registration of the contract for pledge of patent right

Denomination of invention: A Secure Transmission Method of Physical Layer Based on Deep Learning in FDD System

Effective date of registration: 20221027

Granted publication date: 20220318

Pledgee: Hunan Xiangjiang Zhongying Investment Management Co.,Ltd.

Pledgor: HUNAN GUOTIAN ELECTRONIC TECHNOLOGY CO.,LTD.

Registration number: Y2022980019937