CN112688772B - Machine learning superimposed training sequence frame synchronization method - Google Patents
Machine learning superimposed training sequence frame synchronization method Download PDFInfo
- Publication number
- CN112688772B CN112688772B CN202011498196.3A CN202011498196A CN112688772B CN 112688772 B CN112688772 B CN 112688772B CN 202011498196 A CN202011498196 A CN 202011498196A CN 112688772 B CN112688772 B CN 112688772B
- Authority
- CN
- China
- Prior art keywords
- sequence
- frame synchronization
- network
- net
- training
- 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
Links
- 238000012549 training Methods 0.000 title claims abstract description 72
- 238000000034 method Methods 0.000 title claims abstract description 19
- 238000010801 machine learning Methods 0.000 title claims abstract description 9
- 238000005259 measurement Methods 0.000 claims abstract description 26
- 230000005540 biological transmission Effects 0.000 claims abstract description 11
- 238000007781 pre-processing Methods 0.000 claims abstract description 10
- 238000013528 artificial neural network Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 27
- 230000004913 activation Effects 0.000 claims description 11
- 239000000126 substance Substances 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 8
- 230000017105 transposition Effects 0.000 claims description 6
- 108010076504 Protein Sorting Signals Proteins 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 3
- 238000001228 spectrum Methods 0.000 abstract description 4
- 238000004891 communication Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000009977 dual effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Synchronisation In Digital Transmission Systems (AREA)
Abstract
The invention discloses a frame synchronization method for a machine learning superposition training sequence. It comprises the following steps: preprocessing a transmission frame signal generated by a transmitter received by a receiver in a superimposed training sequence mode to obtain a normalized measurement vector of the transmission frame signal; inputting the normalized measurement vector into a trained frame synchronization network FSN-Net to obtain a frame synchronization estimation value thereof, and realizing frame synchronization; inputting a frame synchronization estimation signal obtained according to the frame synchronization estimation value into a trained estimation and equalization sub-network Estequ-Net to obtain an estimation value of a transmission frame signal; eliminating the superimposed training sequence and demodulating the demodulated data in the transmitting frame signal by the estimated value of the transmitting frame signal and the superimposed training sequence mode; the frame synchronization network FSN-Net is constructed based on an ELM network model, and the estimation and equalization sub-network Estequ-Net is constructed based on a deep neural network. The invention can reduce the occupation of frequency spectrum resources and improve the frame synchronization performance, particularly under the condition of nonlinear distortion.
Description
Technical Field
The invention relates to the technical field of wireless communication frame synchronization.
Background
The frame synchronization is one of the important components of the whole wireless communication system, and the performance of the frame synchronization is directly related to the performance of the whole wireless communication system. However, non-linear distortion inevitably exists in the wireless communication system, and the orthogonality of the training sequence is destroyed due to the existence of the non-linear distortion in the conventional frame synchronization method (such as the correlation method), so that the synchronization performance is greatly reduced, and the method is difficult to be applied under the condition of the non-linear distortion.
Disclosure of Invention
The invention aims to provide a frame synchronization method for a machine learning superposition training sequence, which obviously reduces the occupation of frequency spectrum resources and effectively improves the frame synchronization error probability performance under a nonlinear distortion system compared with the traditional correlation synchronization method.
The technical scheme of the invention is as follows:
a machine learning superimposed training sequence frame synchronization method comprises the following steps:
preprocessing a transmission frame signal generated by a transmitter received by a receiver in a superimposed training sequence mode to obtain a normalized measurement vector of the transmission frame signal;
inputting the normalized measurement vector into a trained frame synchronization network FSN-Net to obtain a frame synchronization estimation value thereof, and realizing frame synchronization;
inputting a synchronization signal obtained according to the frame synchronization estimation value into a trained estimation and equalization sub-network EstEqu-Net to obtain an estimation value of a transmission frame signal;
eliminating the superimposed training sequence and demodulating the demodulated data in the transmitting frame signal by the estimated value of the transmitting frame signal and the superimposed training sequence mode;
the frame synchronization network FSN-Net is constructed based on an ELM network model, and the estimation and equalization sub-network Estequ-Net is constructed based on a deep neural network.
According to some preferred embodiments of the present invention, the transmission frame signal is obtained by superimposing a training sequence, as follows:
x＝αs+(1-α)c；
wherein, alpha represents a superposition factor,
a training sequence of length M is represented,
representing a modulated data sequence of length M,
representing an M-dimensional complex field.
According to some preferred embodiments of the invention, the obtaining of the normalized cross-correlation vector comprises:
s21 splicing two frames of same training sequence S into a double training sequence with length of 2M
The following were used:
s22 sequential double training sequence
Middle-intercepting sequence with length of M to generate intercepting sequence
The following were used:
s23 obtaining the truncated sequence through cross-correlation processing
Cross correlation metric Γ with received signal vector y t The following are:
s24 collects M of the cross-correlation metrics Γ t The cross-correlation metric vector γ is constructed as follows:
γ＝[Γ 0 ,Γ 1 ,…,Γ M-1 ] T γ satisfies
Wherein the content of the first and second substances,
representing an M-dimensional real number domain;
s25, normalizing the cross-correlation measurement vector gamma to obtain a normalized cross-correlation measurement vector
The following were used:
the superscript T represents transposition operation, the superscript H represents conjugate transposition operation, and | γ | | | represents the Frobenius norm of the measurement vector γ.
According to some preferred embodiments of the invention, the frame synchronization network FSN-Net comprises:
1 input layer, 1 hidden layer, 1 output layer; the number of nodes of the input layer and the number of nodes of the output layer are equal to the length M of the training sequence, and the number of nodes of the hidden layer is
Wherein the value of m is set according to engineering experience; the activation function of the hidden layer is a sigmoid function.
According to some preferred embodiments of the present invention, the training of the frame synchronization network FSN-Net comprises:
s31 Collection of N t Sequence of M length samples of the received signal
And constructing a sample sequence set
S32 pairs of sample sequence sets
Each signal sequence of
Preprocessing the data to obtain a normalized cross-correlation metric vector sequence in steps S21-S25
Which forms a set of normalized cross-correlation metrics
S33 based on the synchronization offset value τ i ,i＝1,2,…,N t Obtaining a tag sequence T corresponding to the sample sequence through one-hot coding i ,i＝1,2,…,N t Forming a set of tags
Wherein τ is i The label T can be obtained by combining the existing method or equipment according to a statistical channel model or according to an actual scene i ,i＝1,2,…,N t Obtained by one-hot encoding as follows:
s34 generating weights for each normalized cross-correlation metric vector based on Gaussian random distribution
And bias
Cross-correlating the normalized metrics with a metric vector
Forming corresponding label as input of frame synchronization network FSN-Net input layer
The hidden layer output of (2), as follows:
where σ (·) represents an activation function;
s35 harvestSet N t And (3) outputting the hidden layers to form an output matrix H, namely:
wherein the content of the first and second substances,
s36 obtains the output weight β from the hidden layer output matrix H and the label set T by using the following equation:
wherein the content of the first and second substances,
Moore-Penrose pseudoinverse representing H;
s37, model parameters W, b and beta are saved to obtain the frame synchronization network FSN-Net after training.
According to some preferred embodiments of the present invention, the estimating and balancing sub-network EstEqu-Net comprises:
1 input layer, r H Hidden layers, 1 output layer; the number of nodes of the input layer and the number of nodes of the output layer are equal to the length M of the training sequence, and the number of nodes of each layer of the hidden layer is sequentially
l i ≥2，i＝1，...，r H Wherein r is H Not less than 2; the hidden layers all use a Leaky ReLU function as an activation function, and the loss function of the estimation and equalization sub-network EstEqu-Net is a mean square error loss function.
According to some preferred embodiments of the present invention, the frame synchronization estimation signal is transmitted
As training inputs, transmissionsTraining set with frame signal vector x as training label
And training the network, and storing the network model and the parameters after the error is converged to obtain the trained network.
According to some preferred embodiments of the present invention, the obtaining of the demodulated data in the transmitted frame signal estimate comprises:
s51 transmitting frame signal estimated value
Eliminating the superposed sequence to obtain an estimated data sequence
The following were used:
s52 pairs of estimated data sequences
Demodulating to obtain detected data
The invention can efficiently utilize spectrum resources by the Superposition (SC) technology of the training sequences, and can effectively solve the synchronization problem of signals under nonlinear distortion by Machine Learning (ML), such as an ELM network.
The invention effectively combines the advantages of SC and ML technologies, learns the synchronization measurement characteristic of the superimposed training sequence through the ELM network model at the receiving end, thereby accurately estimating the offset position of frame synchronization, and according to the frame starting point, obtaining the emission frame signal through the estimation model established based on the deep neural network, and detecting the data sequence, thereby reducing the occupation of frequency spectrum resources on the basis of improving the frame synchronization error probability performance under the system, particularly the nonlinear distortion system, bringing more implementable schemes for the frame synchronization research, and having great significance.
Drawings
FIG. 1 is a flow chart of the operation of one embodiment of the present invention.
Fig. 2 is a training flowchart of the frame synchronization network FSN-Net according to an embodiment of the present invention.
Detailed Description
The present invention is described in detail with reference to the following embodiments and drawings, but it should be understood that the embodiments and drawings are only for illustrative purposes and are not intended to limit the scope of the present invention. All reasonable variations and combinations that fall within the spirit of the invention are intended to be within the scope of the invention.
According to the technical scheme of the invention, a specific implementation manner is shown in fig. 1, and the method comprises the following steps:
s1 receiver receives the transmitted frame signal generated by the transmitter by the way of superimposed training sequence to form the length M on-line received signal vector
Wherein a frame signal is transmitted
Obtained by means of superimposed training sequences, as follows:
x＝αs+(1-α)c；
wherein, alpha represents a superposition factor which can be set according to engineering experience;
represents a training sequence of length M;
representing a modulated data sequence of length M;
representing an M-dimensional complex field.
S2 preprocesses the on-line received signal vector y to obtain the normalized cross-correlation measurement vector of the on-line received signal vector y and the training sequence
Specifically, the pretreatment comprises:
s21 splicing two frames of same training sequence S into a double training sequence with length of 2M
The following were used:
s22 sequential double training sequence
Middle-intercepting sequence with length of M to generate intercepting sequence
The following:
s23 obtaining the truncated sequence through cross-correlation processing
Cross correlation metric Γ with received signal vector y t The following are:
s24 collects M of the cross-correlation metrics Γ t The cross-correlation metric vector γ is constructed as follows:
γ＝[Γ 0 ,Γ 1 ,…,Γ M-1 ] T γ satisfies
Wherein the content of the first and second substances,
representing an M-dimensional real number domain;
s25, normalizing the cross-correlation measurement vector gamma to obtain a normalized cross-correlation measurement vector
The following:
wherein, the superscript T represents the transposition operation, the superscript H represents the conjugate transposition operation, and | γ | | | represents the Frobenius norm of the measurement vector γ.
S3 cross-correlating the obtained normalized metric with the metric vector
Inputting the trained frame synchronization network FSN-Net to obtain the frame synchronization estimated value
And obtaining a frame synchronization estimation signal
The frame synchronization network FSN-Net can use the following network model:
1 input layer, 1 hidden layer, 1 output layer; the number of nodes of the input layer and the number of nodes of the output layer are equal to the length M of the sequence, and the number of nodes of the hidden layer is
Wherein the value of m is set according to engineering experience; the activation function of the hidden layer is a sigmoid function.
The frame synchronization network FSN-Net can be trained through the process shown in fig. 2, which specifically includes:
s31 Collection of N t Sequence of M length samples of the received signal
And constructing a sample sequence set
S32 pairs of sample sequence sets
Each signal sequence of
Preprocessing the data to obtain a normalized cross-correlation metric vector sequence in steps S21-S25
Which forms a set of normalized cross-correlation metrics
S33 based on the synchronization offset value τ i ,i＝1,2,…,N t Obtaining a tag sequence T corresponding to the sample sequence through one-hot coding i ,i＝1,2,…,N t Forming a set of labels
Wherein τ is i The label T can be obtained by combining the existing method or equipment according to a statistical channel model or according to an actual scene i ,i＝1,2,…,N t Obtained by one-hot encoding as follows:
s34 generating weights for each normalized cross-correlation metric vector based on Gaussian random distribution
And bias
Cross-correlating normalized metrics with a metric vector
Forming corresponding label as input of frame synchronization network FSN-Net input layer
The hidden layer output of (2), as follows:
where σ (·) represents an activation function;
s35 Collection of N t The hidden layers output to form an output matrix H, as follows:
wherein the content of the first and second substances,
s36 obtains the output weight β from the hidden layer output matrix H and the label set T by using the following equation:
wherein, the first and the second end of the pipe are connected with each other,
Moore-Penrose pseudoinverse representing H;
s37, model parameters W, b and beta are saved to obtain the frame synchronization network FSN-Net after training.
S38 collecting received signal sample sequence with length of 2M
Slave sequence
Beginning with the starting point, intercepting the sequence with the length of M to obtain the sample sequence of the online received signal
Y pairs according to steps S21-S25 online Preprocessing the measurement vector to obtain an online normalized measurement cross-correlation measurement vector
Will be provided with
The output vector is learned in a frame synchronization network FSN-Net
The following were used:
s39 finds the index position of the maximum value of the square of the amplitude in the output vector O, i.e., the frame synchronization estimate
The following were used:
s310 root ofData frame synchronization estimation
And receiving a sequence of signal samples
Obtaining a frame synchronization estimation signal
The following were used:
s4 synchronizing signal obtained
Inputting the estimation and equalization sub-network EstEqu-Net after training to obtain the estimation value of the transmission frame signal x
Wherein, estimating and balancing the subnetwork EstEqu-Net can use the following deep neural network:
1 input layer, 2 hidden layers and 1 output layer. The number of nodes of the input layer is M, the number of nodes of the 2 hidden layers is nM, and the number of nodes of the output layer is M;
the value of n can be set according to engineering experience, and 2 hidden layers all adopt ReLU functions as activation functions; the loss function of the network is a mean square error loss function.
The training of the estimation and equalization subnetwork EstEqu-Net comprises:
will be provided with
Training set with transmitted frame signal vector x as training input and training labels
Training the network, and storing a network model and parameters after the error is converged to obtain a trained network;
s5 uses the estimated value of the transmitted frame signal obtained by estimating and equalizing the sub-network EstEqu-Net
Eliminating superimposed training sequence and demodulating demodulation data in transmitting frame signal
Specifically, it may further comprise the steps of:
s51 transmitting frame signal estimated value
Eliminating the superposed sequence to obtain an estimated data sequence
Expressed as:
wherein, alpha represents a superposition factor,
represents a training sequence of length M;
s52 pairs of estimated data sequences
Demodulating to obtain demodulated data
Example 1:
frame synchronization is performed by the process of the detailed embodiment, wherein:
s1 sets M to 64, α to 0.2, N t ＝10 5 ，m＝2，
A receiver receives a transmission frame signal generated by a transmitter in a superimposed training sequence mode to form an online received signal vector y with the length of 64;
a vector x sequence of frame signals is transmitted as follows:
x＝[x 0 ,x 1 ,…,x 63 ] T
training sequence s, as follows:
s＝[s 0 ,s 1 ,…,s 63 ] T
data sequence c, as follows:
c＝[c 0 ,c 1 ,…,c 63 ] T
s2 is based on the setting of S1, and assumes that the received signal vector y is as follows:
y＝[y 0 ,y 1 ,…,y 63 ] T
then the normalized cross-correlation metric vector
The following were used:
dual training sequences
The following were used:
s＝[s 0 ,s 1 ,…,s 63 ,s 0 ,…,s 63 ] T
assuming that t is 3, the generated truncated sequence
As follows
Hypothetical normalized cross-correlation metric vector
The following were used:
s3 according to the setting of S1, the network model of the frame synchronization network FSN-Net is an ELM model, as follows:
1 input layer, 1 hidden layer, 1 output layer; the number of nodes of the input layer and the output layer is 64, and the number of nodes of the hidden layer is 128; the activation function of the hidden layer is a sigmoid function.
The frame synchronization network FSN-Net training process specifically comprises the following steps:
s31 Collection 10 5 A sequence of 64-length received signal samples
And constructing a sample sequence set
S32 pairs of sample sequence sets
Each signal sequence of
Preprocessing the data to obtain a normalized cross-correlation metric vector sequence in steps S21-S25
Which forms a set of normalized cross-correlation metrics "
S33 based on the synchronization offset value τ i ,i＝1,2,…,10 5 Obtaining a tag sequence T corresponding to the sample sequence through one-hot coding i ,i＝1,2,…,10 5 Forming a set of labels
Wherein tau is i The label T can be obtained by combining the existing method or equipment according to a statistical channel model or according to an actual scene i ,i＝1,2,…,N t Obtained by one-hot encoding as follows:
s34 generating weights for each normalized cross-correlation metric vector based on Gaussian random distribution
And bias
Cross-correlating normalized metrics with a metric vector
Forming corresponding label as input of frame synchronization network FSN-Net input layer
The hidden layer output of (2), as follows:
s35 Collection 10 5 The hidden layers output to form an output matrix H, as follows:
s36 obtains the output weight β from the hidden layer output matrix H and the label set T by using the following equation:
wherein the content of the first and second substances,
Moore-Penrose pseudoinverse representing H;
s37, model parameters W, b and beta are saved to obtain the frame synchronization network FSN-Net after training.
S38 collecting a sequence of 128-length received signal samples
Slave sequence
Beginning with the starting point of (1), intercepting the sequence with the length of 64 to obtain the sample sequence of the online received signal
Y pairs according to steps S21-S25 online Preprocessing the measurement vector to obtain an online normalized measurement cross-correlation measurement vector
Will be provided with
The output vector is learned in a frame synchronization network FSN-Net
The following:
s39 finding the index position of the maximum value of the square of the amplitude in the output vector O, i.e. the frame and the frameStep estimation value
The following were used:
s310 estimating value according to frame synchronization
And receiving a sequence of signal samples
Obtaining a frame synchronization estimation signal
The following were used:
s4 estimates and balances the deep neural network that the sub-network EstEqu-Net can use according to the setting of S1, as follows:
1 input layer, 2 hidden layers and 1 output layer. The number of nodes of the input layer is 64, the number of nodes of 2 hidden layers is 128, and the number of nodes of the output layer is 64; wherein, the 2 hidden layers all adopt a ReLU function as an activation function; the loss function of the network is a mean square error loss function.
The training of the estimation and equalization subnetwork EstEqu-Net comprises:
will be provided with
Training set with transmitted frame signal vector x as training input and training labels
Training the network, and storing the network model and parameters after the error is converged to obtain the trained networkComplexing;
s5 is based on the setting of S1 and assumes that the frame signal estimate is transmitted
The sequence is as follows:
hypothesized estimated data sequence
The sequence is as follows:
hypothesis demodulation to obtain demodulated data
Sequence, assume the following:
the above examples are merely preferred embodiments of the present invention, and the scope of the present invention is not limited to the above examples. All technical schemes belonging to the idea of the invention belong to the protection scope of the invention. It should be noted that modifications and embellishments within the scope of the invention may be made by those skilled in the art without departing from the principle of the invention, and such modifications and embellishments should also be considered as within the scope of the invention.
Claims (1)
1. A machine learning superimposed training sequence frame synchronization method is characterized by comprising the following steps:
preprocessing a transmitting frame signal generated by a transmitter in a superimposed training sequence mode and received by a receiver to obtain a normalized measurement vector of the transmitting frame signal;
inputting the normalized measurement vector into a trained frame synchronization network FSN-Net to obtain a frame synchronization estimation value thereof, and realizing frame synchronization;
inputting a frame synchronization estimation signal obtained according to the frame synchronization estimation value into a trained estimation and equalization sub-network Estequ-Net to obtain an estimation value of a transmission frame signal;
eliminating the superimposed training sequence and demodulating the demodulated data in the transmitting frame signal by the estimated value of the transmitting frame signal and the superimposed training sequence mode;
the frame synchronization network FSN-Net is constructed based on an ELM network model, and the estimation and equalization sub-network EstEqu-Net is constructed based on a deep neural network;
the transmission frame signal is obtained by means of a superimposed training sequence, and comprises the following steps:
x＝αs+(1-α)c；
wherein alpha represents a superposition factor, which is set by engineering experience,
a training sequence of length M is represented,
representing a modulated data sequence of length M,
representing an M-dimensional complex field;
the obtaining of the normalized metric vector comprises:
s21 splicing two frames of same training sequence S into a double training sequence with length of 2M
The following:
s22 sequential secondary trainingSequence of
Middle-intercepting the sequence with the length of M to generate an intercepted sequence
The following were used:
s23 obtaining the truncated sequence through cross-correlation processing
Cross-correlation metric Γ with received signal vector y t The following are:
s24 collects M of the cross-correlation metrics Γ t The cross-correlation metric vector γ is constructed as follows:
γ＝[Γ 0 ,Γ 1 ,…,Γ M-1 ] T γ satisfies
Wherein the content of the first and second substances,
representing an M-dimensional real number domain;
s25, normalizing the cross-correlation measurement vector gamma to obtain a normalized cross-correlation measurement vector
The following were used:
the superscript T represents transposition operation, the superscript H represents conjugate transposition operation, and | γ | | | represents Frobenius norm of the measurement vector γ;
the frame synchronization network FSN-Net comprises:
1 input layer, 1 hidden layer, 1 output layer; the number of nodes of the input layer and the number of nodes of the output layer are equal to the length M of the training sequence, and the number of nodes of the hidden layer is
Wherein the value of m is set according to engineering experience; the activation function of the hidden layer is a sigmoid function;
the training of the frame synchronization network FSN-Net comprises the following steps:
s31 Collection of N t Sequence of received signal samples of length M
And constructing a sample sequence set
S32 pairs of sample sequence sets
Each signal sequence of
Preprocessing the data to obtain a normalized cross-correlation metric vector sequence in steps S21-S25
Which forms a set of normalized cross-correlation metrics
S33 based on synchronization offset value tau i ,i＝1,2,…,N t Obtaining a tag sequence T corresponding to the sample sequence through one-hot coding i ,i＝1,2,…,N t Forming a set of labels
Wherein tau is i The label T can be obtained by combining the existing method or equipment according to a statistical channel model or according to an actual scene i ,i＝1,2,…,N t Obtained by one-hot encoding as follows:
s34 generating weights for each normalized cross-correlation metric vector based on Gaussian random distribution
And bias
Cross-correlating normalized metrics with a metric vector
Forming corresponding label as input of frame synchronization network FSN-Net input layer
The hidden layer output of (2), as follows:
where σ (·) represents an activation function;
s35 Collection of N t The hidden layers output to form an output matrix H, namely:
wherein the content of the first and second substances,
s36 obtains the output weight β from the hidden layer output matrix H and the label set T by using the following equation:
wherein the content of the first and second substances,
Moore-Penrose pseudoinverse representing H;
s37, saving the model parameters W, b and beta to obtain a frame synchronization network FSN-Net after training;
the obtaining of the frame synchronization estimation signal comprises:
s38 collecting received signal sample sequence with length of 2M
Slave sequence
Beginning with the starting point, intercepting the sequence with the length of M to obtain the sample sequence of the online received signal
Y pairs according to steps S21-S25 online Preprocessing the measurement vector to obtain an online normalized measurement cross-correlation measurement vector
Will be provided with
The output vector is learned in a frame synchronization network FSN-Net
The following were used:
s39 finds the index position of the maximum value of the square of the amplitude in the output vector O, i.e., the frame synchronization estimate
The following were used:
s310 estimating value according to frame synchronization
And receiving a sequence of signal samples
A frame synchronization estimation signal is obtained as follows:
the estimating and balancing sub-network EstEqu-Net comprises:
1 input layer, r H Hidden layers, 1 output layer; the number of nodes of the input layer and the number of nodes of the output layer are equal to the length M of the training sequence, and the number of nodes of each layer of the hidden layer is l in sequence 1 M D ,l 2 M D ,...,
l i ≥2,i＝1,...,r H Wherein r is H Not less than 2; the hidden layer takes an escape ReLU function as an activation function, and a loss function of the estimation and equalization sub-network EstEqu-Net is a mean square error loss function;
the training of the estimation and equalization subnetwork EstEqu-Net comprises:
estimating the frame synchronization
Training set with transmitted frame signal vector x as training input and training labels
Training the network, and storing a network model and parameters after the error is converged to obtain a trained network;
transmitting frame signal estimated value obtained by utilizing estimating and equalizing sub-network Estequ-Net
Eliminating superimposed training sequence and demodulating demodulation data in transmitting frame signal
The method comprises the following steps:
s51 transmitting frame signal estimated value
Eliminating the superposed sequence to obtain an estimated data sequence
The following were used:
wherein α represents a superposition factor，
Represents a training sequence of length M;
s52 pairs of estimated data sequences
Demodulating to obtain demodulated data
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011498196.3A CN112688772B (en) | 2020-12-17 | 2020-12-17 | Machine learning superimposed training sequence frame synchronization method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011498196.3A CN112688772B (en) | 2020-12-17 | 2020-12-17 | Machine learning superimposed training sequence frame synchronization method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112688772A CN112688772A (en) | 2021-04-20 |
| CN112688772B true CN112688772B (en) | 2022-08-26 |
Family
ID=75448856
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011498196.3A Active CN112688772B (en) | 2020-12-17 | 2020-12-17 | Machine learning superimposed training sequence frame synchronization method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112688772B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114096000B (en) * | 2021-11-18 | 2023-06-23 | 西华大学 | Combined frame synchronization and channel estimation method based on machine learning |
| CN114157544B (en) * | 2021-12-07 | 2023-04-07 | 中南大学 | Frame synchronization method, device and medium based on convolutional neural network |
| CN117295149B (en) * | 2023-11-23 | 2024-01-30 | 西华大学 | Frame synchronization method and system based on low-complexity ELM assistance |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101292481A (en) * | 2005-09-06 | 2008-10-22 | 皇家飞利浦电子股份有限公司 | Method and apparatus for estimating channel based on implicit training sequence |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102291360A (en) * | 2011-09-07 | 2011-12-21 | 西南石油大学 | Superimposed training sequence based optical OFDM (Orthogonal Frequency Division Multiplexing) system and frame synchronization method thereof |
| TWI446770B (en) * | 2012-01-20 | 2014-07-21 | Nat Univ Tsing Hua | Communication system having data-dependent superimposed training mechanism and communication method thereof |
| CN110830112A (en) * | 2019-10-16 | 2020-02-21 | 青岛海信电器股份有限公司 | Visible light communication method and device |
| CN111970078B (en) * | 2020-08-14 | 2022-08-16 | 西华大学 | Frame synchronization method for nonlinear distortion scene |
-
2020
- 2020-12-17 CN CN202011498196.3A patent/CN112688772B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101292481A (en) * | 2005-09-06 | 2008-10-22 | 皇家飞利浦电子股份有限公司 | Method and apparatus for estimating channel based on implicit training sequence |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112688772A (en) | 2021-04-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112688772B (en) | Machine learning superimposed training sequence frame synchronization method | |
| Gao et al. | ComNet: Combination of deep learning and expert knowledge in OFDM receivers | |
| CN111404849B (en) | OFDM channel estimation and signal detection method based on deep learning | |
| CN109687897B (en) | Superposition CSI feedback method based on deep learning large-scale MIMO system | |
| JP2001517399A (en) | Self-synchronous equalization method and system | |
| CN108540419A (en) | A kind of OFDM detection methods of the anti-inter-sub-carrier interference based on deep learning | |
| CN109246039A (en) | A kind of Soft Inform ation iteration receiving method based on two-way time domain equalization | |
| CN113395225B (en) | Universal intelligent processing method and device for directly receiving communication signal waveform to bit | |
| CN111970078B (en) | Frame synchronization method for nonlinear distortion scene | |
| CN110971457B (en) | Time synchronization method based on ELM | |
| CN112215335B (en) | System detection method based on deep learning | |
| CN102111360A (en) | Algorithm for dynamically switching channel equalization based on real-time signal-to-noise ratio estimation | |
| CN112598072A (en) | Equalization method of improved Volterra filter based on weight coefficient migration of SVM training | |
| CN101873295B (en) | Signal processing method and device as well as signal receiving method and receiving machine | |
| Smith et al. | A communication channel density estimating generative adversarial network | |
| CN106656881A (en) | Adaptive blind equalization method based on deviation compensation | |
| CN113381953A (en) | Channel estimation method of extreme learning machine based on reconfigurable intelligent surface assistance | |
| Wang et al. | Online LSTM-based channel estimation for HF MIMO SC-FDE system | |
| CN114499601B (en) | Large-scale MIMO signal detection method based on deep learning | |
| Ponnaluru et al. | RETRACTED ARTICLE: Deep learning for estimating the channel in orthogonal frequency division multiplexing systems | |
| CN112491754B (en) | Channel estimation and signal detection method based on DDST and deep learning | |
| CN104868962B (en) | Frequency spectrum detecting method and device based on compressed sensing | |
| CN114513394A (en) | Attention machine drawing neural network-based signal modulation format identification method, system and device and storage medium | |
| CN110944002B (en) | Physical layer authentication method based on exponential average data enhancement | |
| CN101651643A (en) | Blind equalization method for wavelet neural network based on space diversity |
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 | ||
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20210420 Assignee: Suining Feidian Cultural Communication Co.,Ltd. Assignor: XIHUA University Contract record no.: X2023510000027 Denomination of invention: A Machine Learning Overlay Training Sequence Frame Synchronization Method Granted publication date: 20220826 License type: Common License Record date: 20231129 |
|
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20210420 Assignee: Chengdu Suyouyun Information Technology Co.,Ltd. Assignor: XIHUA University Contract record no.: X2023510000030 Denomination of invention: A Machine Learning Overlay Training Sequence Frame Synchronization Method Granted publication date: 20220826 License type: Common License Record date: 20231201 |
|
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20210420 Assignee: Chengdu Yingling Feifan Technology Co.,Ltd. Assignor: XIHUA University Contract record no.: X2023510000032 Denomination of invention: A Machine Learning Overlay Training Sequence Frame Synchronization Method Granted publication date: 20220826 License type: Common License Record date: 20231212 Application publication date: 20210420 Assignee: Sichuan Shenglongxing Technology Co.,Ltd. Assignor: XIHUA University Contract record no.: X2023510000031 Denomination of invention: A Machine Learning Overlay Training Sequence Frame Synchronization Method Granted publication date: 20220826 License type: Common License Record date: 20231211 |