CN112906601B - Signal-to-noise ratio estimation method for multiple signals - Google Patents
Signal-to-noise ratio estimation method for multiple signals Download PDFInfo
- Publication number
- CN112906601B CN112906601B CN202110239443.6A CN202110239443A CN112906601B CN 112906601 B CN112906601 B CN 112906601B CN 202110239443 A CN202110239443 A CN 202110239443A CN 112906601 B CN112906601 B CN 112906601B
- Authority
- CN
- China
- Prior art keywords
- signal
- data
- training
- feature
- noise ratio
- 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
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000012549 training Methods 0.000 claims abstract description 103
- 238000000605 extraction Methods 0.000 claims abstract description 39
- 238000012545 processing Methods 0.000 claims abstract description 30
- 238000012360 testing method Methods 0.000 claims description 56
- 230000004913 activation Effects 0.000 claims description 18
- 102100026758 Serine/threonine-protein kinase 16 Human genes 0.000 claims description 16
- 101710184778 Serine/threonine-protein kinase 16 Proteins 0.000 claims description 16
- 238000011176 pooling Methods 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 10
- 210000002569 neuron Anatomy 0.000 claims description 9
- 230000005540 biological transmission Effects 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 238000013527 convolutional neural network Methods 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 6
- 238000011160 research Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 3
- 238000013178 mathematical model Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 230000004083 survival effect Effects 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims 1
- 238000004891 communication Methods 0.000 description 11
- 238000005516 engineering process Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 238000013461 design Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000012417 linear regression Methods 0.000 description 2
- 241000282414 Homo sapiens Species 0.000 description 1
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000001568 sexual effect Effects 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- 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
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
Abstract
The invention discloses a signal-to-noise ratio estimation method for various signals, which comprises the steps of generating first signal training data and processing the first signal training data to obtain second signal training data; extracting time domain characteristics of the second signal training data and processing to obtain second signal training characteristic data; building an original signal feature extraction model and training by adopting second signal training data to obtain second signal training data features and a pre-training signal feature extraction model; fusing the second signal training characteristic data and the second signal training data characteristic to obtain a first combined characteristic; performing ridge regression training on the first combined features to obtain a ridge regression signal-to-noise ratio estimation model; and performing signal-to-noise ratio estimation by using a ridge regression signal-to-noise ratio estimation model. The invention can effectively improve the prediction precision, enlarge the range of the predicted signal-to-noise ratio, increase the types of the predicted signals, prevent overfitting, and has the advantages of high reliability, good stability, wider applicable signal-to-noise ratio range, more signal types and better estimation performance under the lower signal-to-noise ratio.
Description
Technical Field
The invention belongs to the field of communication measurement, and particularly relates to a signal-to-noise ratio estimation method for various signals.
Background
The signal-to-noise ratio, i.e. the ratio of the signal power to the noise power, is an important index for measuring the signal quality and the quality of the circuit design of the communication system. The modern society can not use the communication technology, and the development of the communication technology greatly promotes the improvement of the living standard of human beings. In the conventional communication technology, the signal-to-noise ratio estimation is an important subject, and can help people to reasonably design or select appropriate communication components so as to extract signals with better quality. The good signal-to-noise ratio estimation model or method can help people to better control the signal transmission power and also help communication companies to obtain better balance between resource control and signal service. In the field of quantum communication, bit error rate is an important criterion for judging whether communication is stolen, and estimation of the bit error rate also requires prior knowledge of signal-to-noise ratio estimation.
The existing snr estimation techniques, such as maximum likelihood snr estimation, M2M4 estimator, etc., have an estimated snr range over 0dB in general, and have limited applicable signal types (only for one type of signal), such as M2M4 estimator, which is only effective for low-order PSK modulation. Therefore, the application range of the existing signal-to-noise ratio estimation technology is narrow, and the universality is poor; and with the development of quantum communication technology, the existing signal-to-noise ratio estimation technology is not suitable for the field of quantum communication.
Disclosure of Invention
The invention aims to provide a signal-to-noise ratio estimation method which can estimate signal-to-noise ratios in a wide range, is applicable to various signal types and has better estimation performance under the condition of lower signal-to-noise ratio.
The signal-to-noise ratio estimation method for various signals provided by the invention comprises the following steps:
s1, generating first signal training data D1 according to a signal and a channel model to be researched;
s2, processing the first signal training data D1 acquired in the step S1 to obtain second signal training data D2;
s3, extracting and processing time domain characteristics of second signal training data D2 to obtain second signal training characteristic data F1;
s4, building an original signal feature extraction model;
s5, training the original signal feature extraction model built in the step S4 by taking the second signal training data D2 as a training sample, so as to obtain second signal training data features F2 and a pre-training signal feature extraction model;
s6, fusing the second signal training feature data F1 obtained in the step S3 and the second signal training data feature F2 obtained in the step S5 to obtain a first combined feature;
s7, performing ridge regression training on the first combined features obtained in the step S6 to obtain a ridge regression signal-to-noise ratio estimation model;
and S8, estimating the signal-to-noise ratio of the target signal by using the ridge regression signal-to-noise ratio estimation model obtained in the step S7.
The signal-to-noise ratio estimation method for various signals further comprises the following steps:
s9, generating and processing first signal test data T1 to obtain second signal test data T2;
s10, extracting time domain characteristics of the second signal test data T2 and processing the time domain characteristics to obtain second signal test characteristic data FT 1;
s11, extracting the characteristics of the second signal test data T2 obtained in the step S9 by adopting the pre-training signal characteristic extraction model obtained in the step S5, so that second signal test data characteristics FT2 are obtained;
s12, fusing the second signal test characteristic data FT1 obtained in the step S10 and the second signal test data characteristic FT2 obtained in the step S11 to obtain a second combined characteristic;
and S13, testing the second combined features acquired in the step S12 by using the ridge regression signal-to-noise ratio estimation model acquired in the step S7, so as to evaluate the performance of the ridge regression signal-to-noise ratio estimation model.
In step S1, first signal training data D1 is generated according to the signal and channel model to be studied, specifically, the following steps are adopted to generate data:
A. generating data by adopting a Matlab writing program;
B. for the MPSK/MQAM modulation signal, the set modulation orders M comprise 2, 4, 8, 16, 32 and 64; for Gaussian pulse signals, a modulation order does not need to be set;
C. setting an estimated SNR range according to the interested research range;
D. symbol upsampling frequency NssSetting the frequency to be 2 times or more of the signal frequency;
E. setting the number of one-time transmission signals NsymThe number of samples, numbits, corresponding to each SNR;
F. and setting the order orderRcos of the forming filter.
In step S2, the processing of the first signal training data D1 obtained in step S1 is specifically to perform 0-1 normalization processing on the first signal training data obtained in step S1.
The step S3 is to extract and process the time domain feature of the second signal training data D2, so as to obtain second signal training feature data F1, specifically, the following steps are adopted for extraction and processing:
a. for a gaussian pulse signal, the extracted time domain features include: a maximum, a minimum, a mean, a peak-to-peak, a rectified mean, a variance, a standard deviation, a kurtosis, a root mean square, a form factor, a peak factor, a kurtosis factor, a pulse factor, and a margin factor;
b. for MPSK/MQAM type modulation signals, the signals are divided into real parts and imaginary parts, and time domain characteristics which are respectively extracted aiming at the real part data and the imaginary part data comprise: a maximum, a minimum, a mean, a peak-to-peak, a rectified mean, a variance, a standard deviation, a kurtosis, a root mean square, a form factor, a peak factor, a kurtosis factor, a pulse factor, and a margin factor;
c. and (c) performing 0-1 normalization processing on the time domain features extracted in the step a and the step b to obtain final second signal training feature data F1.
Step S4, building an original signal feature extraction model, specifically building an original signal feature extraction model by the following steps:
a convolutional neural network model is adopted as a feature extraction model;
the network structure of the model comprises a convolution layer, a pooling layer, a flattening layer and a full-connection layer;
the convolution layer adopts one-dimensional convolution to carry out data operation, and the mathematical model is
In the formula OiFor the i-th output of the network, WiIs a weight matrix of the i-th layer, biIs the bias of the ith layer;
representing a convolution operation; all convolution layer activation function uses lineA sexual activation function;
the pooling layer is used for secondary feature extraction or fusion; the pooling layer reduces the dimension of the value in the receptive field to 1 value, and then the next pooling value is obtained through the movement of fixed step length step by step;
the flattening layer is used for matching with the full connection layer to perform fitting network training and reducing the characteristics into 1-dimensional characteristics;
the full connection layer is used for carrying out regression fitting; the fully-connected layer comprises three fully-connected sublayers, the first fully-connected sublayer comprises 128 neurons, the second fully-connected sublayer comprises 32 neurons, and the third fully-connected sublayer comprises 1 neuron; the activation functions of the first fully-connected sublayer and the second fully-connected sublayer are linear rectification functions, and the activation function of the third fully-connected sublayer is
The activation functions of all the convolution layers adopt linear activation functions, specifically adopt linear rectification functions Relu
The activation function of the first fully-connected sublayer and the second fully-connected sublayer is a linear rectification function, and specifically adopts a Relu function
The step S5 of obtaining the second signal training data feature F2 is to obtain the feature extracted by the second fully-connected sub-layer in the pre-training signal feature extraction model, and use the feature as the second signal training data feature.
In step S6, the first combined feature is obtained by fusing the second signal training feature data F1 obtained in step S3 and the second signal training data feature F2 obtained in step S5, specifically, the second signal training feature data F1 obtained in step S3 and the second signal training data feature F2 obtained in step S5 are fused in a manner of F1+ F2.
Step S7Performing ridge regression training on the first combined features obtained in the step S6 to obtain a ridge regression signal-to-noise ratio estimation model, specifically selecting a ridge regression model to perform ridge regression training; the target function of the ridge regression model is argmin [ L (w) + lambda P (w)]Wherein L (w) is a loss function; λ is a regularization factor; argmin () is the value of the argument that minimizes the parenthetical inner formula; p (w) is a penalty term; regularization is L1 regression and L2 regression, with P (w) | | w | | survival in the L1 regression1In the L2 regression
The ridge regression used the L2 regression, and the Lasso regression used the L1 regression.
The step S9 of generating and processing the first signal test data T1 specifically includes the following steps:
(1) generating data by adopting a Matlab writing program;
(2) for the MPSK/MQAM modulation signal, the set modulation orders M comprise 2, 4, 8, 16, 32 and 64; for Gaussian pulse signals, a modulation order does not need to be set;
(3) setting an estimated SNR range according to the interested research range;
(4) symbol up-sampling frequency NssSetting the frequency to be 2 times or more of the signal frequency;
(5) setting the number of one-time transmission signals NsymThe number of samples, numbits, corresponding to each SNR;
(6) setting the order orderRcos of the forming filter;
(7) and carrying out normalization processing on the generated data.
The step S10 is to extract and process the time domain feature of the second signal test data T2, so as to obtain second signal test feature data FT1, specifically, the following steps are adopted for extraction and processing:
a. for a gaussian pulse signal, the extracted time domain features include: a maximum, a minimum, a mean, a peak-to-peak, a rectified mean, a variance, a standard deviation, a kurtosis, a root mean square, a form factor, a peak factor, a kurtosis factor, a pulse factor, and a margin factor;
b. for MPSK/MQAM modulated signals, the signals are divided into real parts and imaginary parts, and time domain characteristics which are respectively extracted according to real part data and data which need to be complemented comprise: maximum, minimum, mean, peak-to-peak, rectified mean, variance, standard deviation, kurtosis, root mean square, form factor, peak factor, kurtosis factor, impulse factor, and margin factor;
c. and c, performing normalization processing on the time domain features extracted in the step a and the step b to obtain final second signal test feature data FT 1.
In step S11, the pre-training signal feature extraction model obtained in step S5 is used to extract features of the second signal test data T2 obtained in step S9, so as to obtain a second signal test data feature FT2, specifically, the second signal test data T2 is input to the pre-training signal feature extraction model, and features extracted by the second full-link sublayer in the pre-training signal feature extraction model are obtained and used as the second signal test data features.
Step S12, which is to fuse the second signal test feature data FT1 obtained in step S10 and the second signal test data feature FT2 obtained in step S11, so as to obtain a second combined feature, specifically, fuse the second signal test feature data FT1 obtained in step S10 and the second signal test data feature FT2 obtained in step S11 in a manner of FT1+ FT2, so as to obtain a second combined feature.
The performance evaluation of the ridge regression signal-to-noise ratio estimation model described in step S13 is specifically performed by using Bias, normalized MEAN, variance VAR, normalized Bias NBias, and normalized variance NMSE as evaluation indexes to evaluate the ridge regression signal-to-noise ratio estimation model.
The signal-to-noise ratio estimation method for various signals provided by the invention innovatively applies convolutional neural network and ridge regression to the prediction of the signal-to-noise ratio, and provides a CNN + ridge regression model, so that the method can effectively improve the prediction precision, enlarge the range of the predicted signal-to-noise ratio, increase the types of predicted signals, prevent overfitting, and has the characteristics of high reliability and good stability, wider range of the estimated signal-to-noise ratio, more applicable signal types and better estimation performance under the lower signal-to-noise ratio.
Drawings
FIG. 1 is a schematic process flow diagram of the process of the present invention.
FIG. 2 is a schematic view showing a broken line comparing an estimated value and a true value when the method of the present invention estimates MPSK/MQAM signals.
FIG. 3 is a schematic diagram of the deviation broken line after the SNR of different modulation signals is estimated by the method of the present invention.
FIG. 4 is a schematic diagram of variance broken lines after the SNR of different modulation signals is estimated by the method of the present invention.
FIG. 5 is a schematic graph showing a comparison between an estimated value and a true value when the Gaussian pulse signal is estimated according to the method of the present invention.
FIG. 6 is a polygonal line diagram illustrating the deviation between the estimated value and the true value when the Gaussian pulse signal is estimated by the method of the present invention.
Detailed Description
FIG. 1 is a schematic flow chart of the method of the present invention: the signal-to-noise ratio estimation method for various signals provided by the invention comprises the following steps:
s1, generating first signal training data D1 according to a signal and a channel model to be researched; specifically, the following steps are adopted to generate data:
A. generating data by adopting a Matlab writing program;
B. for the MPSK/MQAM modulation signal, the set modulation orders M comprise 2, 4, 8, 16, 32 and 64; for Gaussian pulse signals, a modulation order does not need to be set;
C. setting an estimated SNR range according to the interested research range; for example, it can be set to-20 dB to 20 dB;
D. symbol up-sampling frequency NssSetting the frequency of the signal to be 2 times or more of the frequency of the signal so as to accord with the Nyquist sampling theorem;
E. setting the number of one-time transmission signals NsymThe number of samples, numbits, corresponding to each SNR;
F. setting the order orderRcos of the formed filter; except for Gaussian pulse signals, the signals are required to be set, and after up-sampling, the baseband signals are required to be subjected to pulse forming to be transmitted into a channel for transmission;
s2, processing the first signal training data D1 acquired in the step S1 to obtain second signal training data D2; specifically, 0-1 normalization processing is performed on the first signal training data acquired in step S1;
thereby normalizing the training data to a number between [ -1,1 ];
s3, extracting and processing time domain characteristics of second signal training data D2 to obtain second signal training characteristic data F1; the method specifically comprises the following steps of:
a. for a gaussian pulse signal, the extracted time domain features include: a maximum, a minimum, a mean, a peak-to-peak, a rectified mean, a variance, a standard deviation, a kurtosis, a root mean square, a form factor, a peak factor, a kurtosis factor, a pulse factor, and a margin factor;
b. for MPSK/MQAM type modulation signals, the signals are divided into real parts and imaginary parts, and time domain characteristics which are respectively extracted aiming at the real part data and the imaginary part data comprise: a maximum, a minimum, a mean, a peak-to-peak, a rectified mean, a variance, a standard deviation, a kurtosis, a root mean square, a form factor, a peak factor, a kurtosis factor, a pulse factor, and a margin factor; meanwhile, according to the characteristics of the MPSK/MQAM signal, the statistical characteristics of the real part and the imaginary part of the MPSK/MQAM signal are consistent, so that only the real part characteristic can be selected as the subsequent training characteristic;
c. b, performing 0-1 normalization processing on the time domain features extracted in the step a and the step b to obtain final second signal training feature data F1;
s4, building an original signal feature extraction model; specifically, an original signal feature extraction model is built by adopting the following steps:
a convolutional neural network model is adopted as a feature extraction model;
the network structure of the model comprises a convolution layer, a pooling layer, a flattening layer and a full-connection layer;
the convolution layer adopts one-dimensional convolution to carry out data operation, and the mathematical model is
In the formula OiFor the i-th output of the network, WiIs a weight matrix of the i-th layer, biIs the bias of the ith layer;
representing a convolution operation; the activation functions of all the convolution layers adopt linear activation functions; in particular, a linear rectification function Relu function is adopted
The pooling layer is used for performing secondary feature extraction or fusion; the pooling layer reduces the dimension of the value in the receptive field to 1 value, and then the next pooling value is obtained through the movement of fixed step length step by step;
the flattening layer is used for matching with the full connection layer to perform fitting network training and reducing the characteristics into 1-dimensional characteristics;
the full connection layer is used for carrying out regression fitting; the fully-connected layer comprises three fully-connected sublayers, the first fully-connected sublayer comprises 128 neurons, the second fully-connected sublayer comprises 32 neurons, and the third fully-connected sublayer comprises 1 neuron; the activation functions of the first and second fully-connected sublayers are linear rectification functions, specifically adopting Relu function
The activation function of the third fully-connected sublayer is
S5, training the original signal feature extraction model built in the step S4 by taking the second signal training data D2 as a training sample, so as to obtain second signal training data features F2 and a pre-training signal feature extraction model; specifically, the features extracted by the second fully-connected sublayer in the pre-training signal feature extraction model are obtained and used as the features of the second signal training data;
s6, fusing the second signal training feature data F1 obtained in the step S3 and the second signal training data feature F2 obtained in the step S5 to obtain a first combined feature; specifically, the second signal training feature data F1 obtained in step S3 and the second signal training feature data F2 obtained in step S5 are fused in a manner of F1+ F2;
s7, performing ridge regression training on the first combined features obtained in the step S6 to obtain a ridge regression signal-to-noise ratio estimation model; specifically, a ridge regression model is selected to carry out ridge regression training; the target function of the ridge regression model is argmin [ L (w) + lambda P (w)]Wherein L (w) is a loss function; λ is a regularization factor; argmin () is the value of the argument that minimizes the parenthetical inner formula; p (w) is a penalty term; regularization is divided into L1 regression and L2 regression, and p (w) | | w | | survival in L1 regression1In the L2 regression
The ridge regression used the L2 regression, and the Lasso regression used the L1 regression;
the essence of the method is the optimization of a least square method, the method has strong anti-noise capability, multiple collinearity possibly exists among the extracted characteristics, the variance of a parameter estimation value is increased due to the multiple collinearity, the multiple collinearity cannot be fitted by linear regression, and compared with the linear regression, the ridge regression is more suitable for the fitting of the data with the multiple collinearity characteristics, and a pre-trained ridge regression model is obtained after training;
s8, estimating the signal-to-noise ratio of the target signal by adopting the ridge regression signal-to-noise ratio estimation model obtained in the step S7;
s9, generating and processing first signal test data T1 to obtain second signal test data T2; specifically, the following steps are adopted to generate and process data:
(1) generating data by adopting a Matlab writing program;
(2) for the MPSK/MQAM modulation signal, the set modulation orders M comprise 2, 4, 8, 16, 32 and 64; for Gaussian pulse signals, the modulation order does not need to be set;
(3) setting an estimated SNR range according to the interested research range;
(4) symbol up-sampling frequency NssSetting the frequency to be 2 times or more of the signal frequency;
(5) setting the number of one-time transmission signals NsymThe number of samples, numbits, corresponding to each SNR;
(6) setting the order orderRcos of the formed filter;
(7) carrying out normalization processing on the generated data;
s10, extracting and processing time domain characteristics of the second signal test data T2 to obtain second signal test characteristic data FT 1; the method specifically comprises the following steps of:
a. for a gaussian pulse signal, the extracted time domain features include: a maximum, a minimum, a mean, a peak-to-peak, a rectified mean, a variance, a standard deviation, a kurtosis, a root mean square, a form factor, a peak factor, a kurtosis factor, a pulse factor, and a margin factor;
b. for MPSK/MQAM modulated signals, the signals are divided into real parts and imaginary parts, and time domain characteristics which are respectively extracted according to real part data and data which need to be complemented comprise: a maximum, a minimum, a mean, a peak-to-peak, a rectified mean, a variance, a standard deviation, a kurtosis, a root mean square, a form factor, a peak factor, a kurtosis factor, a pulse factor, and a margin factor;
c. b, normalizing the time domain features extracted in the step a and the step b to obtain final second signal test feature data FT 1;
s11, extracting the characteristics of the second signal test data T2 obtained in the step S9 by adopting the pre-training signal characteristic extraction model obtained in the step S5, so that second signal test data characteristics FT2 are obtained; inputting second signal test data T2 into a pre-training signal feature extraction model, acquiring features extracted by a second full-connection sublayer in the pre-training signal feature extraction model, and taking the features as second signal test data features;
s12, fusing the second signal test characteristic data FT1 obtained in the step S10 and the second signal test data characteristic FT2 obtained in the step S11 to obtain a second combined characteristic; specifically, the second signal test characteristic data FT1 obtained in the step S10 and the second signal test characteristic data FT2 obtained in the step S11 are fused in a mode of FT1+ FT2, so that a second combined characteristic is obtained;
s13, testing the second combined features obtained in the step S12 by using the ridge regression signal-to-noise ratio estimation model obtained in the step S7, and thus performing performance evaluation on the ridge regression signal-to-noise ratio estimation model; specifically, the deviation Bias, the normalized MEAN, the variance VAR, the normalized deviation NBias and the normalized variance NMSE are used as evaluation indexes to evaluate the ridge regression signal-to-noise ratio estimation model.
FIG. 2 shows a line graph of the true and predicted values of the M-PSK/QAM signal predicted value for the proposed model estimation; fig. 3 shows a deviation line graph of the proposed model for predicted values of snr of different modulation signals, and fig. 4 shows a deviation line graph of the proposed model for predicted values of snr of different modulation signals, it can be seen that the conventional M2M4 estimator can only estimate snr of more than minus 8dB and can only estimate MPSK signals, and loses the prediction capability for QAM signals, whereas the CNN + ridge regression model proposed in this patent can estimate both the snr of PSK signals and the snr of QAM signals. And when the signal is more than-15 dB, the estimation deviation is only about 2dB, 8psk signals are removed, and the rest signals are all within 2dB, so that the method provided by the patent has good low signal-to-noise ratio prediction capability.
Fig. 5 shows a comparison graph of the true value and the predicted value of the gaussian pulse signal estimated by the proposed model, and fig. 6 shows a deviation line graph of the predicted value of the gaussian pulse signal estimated by the proposed model, it can be seen that the model proposed by the present patent has a stronger prediction capability than that of predicting a modulation signal, within a range of-20 dB to 20dB, the maximum absolute error is only 3dB, the predicted amplitude absolute error is smaller, and the validity of the present patent is further verified because a smaller signal-to-noise ratio value corresponds to a smaller amplitude.
Claims (10)
1. A method for signal-to-noise ratio estimation of a plurality of signals, comprising the steps of:
s1, generating first signal training data according to a signal to be researched and a channel model;
s2, processing the first signal training data obtained in the step S1 to obtain second signal training data;
s3, extracting and processing time domain characteristics of the second signal training data to obtain second signal training characteristic data;
s4, building an original signal feature extraction model;
s5, training the original signal feature extraction model built in the step S4 by taking the second signal training data as a training sample so as to obtain second signal training data features and a pre-training signal feature extraction model;
s6, fusing the second signal training characteristic data obtained in the step S3 and the second signal training data obtained in the step S5 to obtain a first combined characteristic;
s7, performing ridge regression training on the first combined features obtained in the step S6 to obtain a ridge regression signal-to-noise ratio estimation model;
and S8, estimating the signal-to-noise ratio of the target signal by using the ridge regression signal-to-noise ratio estimation model obtained in the step S7.
2. The method of claim 1, further comprising the steps of:
s9, generating and processing first signal test data to obtain second signal test data;
s10, extracting time domain characteristics of the second signal test data and processing the time domain characteristics to obtain second signal test characteristic data;
s11, extracting the characteristics of the second signal test data obtained in the step S9 by adopting the pre-training signal characteristic extraction model obtained in the step S5 so as to obtain the characteristics of the second signal test data;
s12, fusing the second signal test characteristic data obtained in the step S10 and the second signal test data characteristic obtained in the step S11 to obtain a second combined characteristic;
and S13, testing the second combined features acquired in the step S12 by using the ridge regression signal-to-noise ratio estimation model acquired in the step S7, so as to evaluate the performance of the ridge regression signal-to-noise ratio estimation model.
3. The method according to claim 1 or 2, wherein the step S1 generates the first signal training data according to the signal and channel model to be studied, specifically, the following steps are adopted to generate the data:
A. generating data by adopting a Matlab writing program;
B. for the MPSK/MQAM modulation signal, the set modulation orders M comprise 2, 4, 8, 16, 32 and 64; for Gaussian pulse signals, a modulation order does not need to be set;
C. setting an estimated SNR range according to the interested research range;
D. symbol upsampling frequency NssSetting the frequency to be 2 times or more of the signal frequency;
E. setting the number of one-time transmission signals NsymThe number of samples, numbits, corresponding to each SNR;
F. and setting the order orderRcos of the forming filter.
4. The method as claimed in claim 3, wherein the step S3 extracts and processes the time domain feature of the second signal training data D2 to obtain the second signal training feature data F1, specifically by the steps of:
a. for a gaussian pulse signal, the extracted time domain features include: a maximum, a minimum, a mean, a peak-to-peak, a rectified mean, a variance, a standard deviation, a kurtosis, a root mean square, a form factor, a peak factor, a kurtosis factor, a pulse factor, and a margin factor;
b. for MPSK/MQAM type modulation signals, the signals are divided into real parts and imaginary parts, and time domain characteristics which are respectively extracted aiming at the real part data and the imaginary part data comprise: a maximum, a minimum, a mean, a peak-to-peak, a rectified mean, a variance, a standard deviation, a kurtosis, a root mean square, a form factor, a peak factor, a kurtosis factor, a pulse factor, and a margin factor;
c. and (c) performing 0-1 normalization processing on the time domain features extracted in the step a and the step b to obtain final second signal training feature data F1.
5. The method according to claim 4, wherein the original signal feature extraction model is built in step S4, specifically, the original signal feature extraction model is built by the following steps:
a convolutional neural network model is adopted as a feature extraction model;
the network structure of the model comprises a convolution layer, a pooling layer, a flattening layer and a full-connection layer;
the convolution layer adopts one-dimensional convolution to carry out data operation, and the mathematical model is
In the formula OiFor the i-th output of the network, WiWeight matrix for i-th layer, biIs the bias of the ith layer;
representing a convolution operation; the activation functions of all the convolution layers adopt linear activation functions;
the pooling layer is used for secondary feature extraction or fusion; the pooling layer reduces the dimension of the value in the receptive field to 1 value, and then the next pooling value is obtained through the movement of fixed step length step by step;
the flattening layer is used for matching with the full connection layer to perform fitting network training and reducing the characteristics into 1-dimensional characteristics;
the full connection layer is used for carrying out regression fitting; the fully-connected layer comprises three fully-connected sublayers, the first fully-connected sublayer comprises 128 neurons, the second fully-connected sublayer comprises 32 neurons, and the third fully-connected sublayer comprises 1 neuron; first of allThe activation functions of the fully-connected sublayer and the second fully-connected sublayer are linear rectification functions, and the activation function of the third fully-connected sublayer is
6. Method for signal-to-noise ratio estimation of multiple signals according to claim 5, characterized in that the activation functions of all convolutional layers are linear activation functions, in particular linear rectification functions Relu
The activation function of the first fully-connected sublayer and the second fully-connected sublayer is a linear rectification function, and specifically adopts a Relu function
7. The method of claim 6, wherein the step S5 is performed to obtain a second signal training data feature F2, specifically to obtain the feature extracted by the second fully-connected sub-layer in the pre-training signal feature extraction model, and to use the feature as the second signal training data feature.
8. The method of claim 7, wherein the step S6 is performed by fusing the second signal training characteristic data F1 obtained in the step S3 and the second signal training data characteristic F2 obtained in the step S5 to obtain a first combined characteristic, and specifically, the second signal training characteristic data F1 obtained in the step S3 and the second signal training data characteristic F2 obtained in the step S5 are fused in a manner of F1+ F2.
9. The method of claim 8, wherein the step S7 is performed on the first combined signature obtained in the step S6Performing ridge regression training to obtain a ridge regression signal-to-noise ratio estimation model, specifically selecting a ridge regression model to perform ridge regression training; the target function of the ridge regression model is argmin [ L (w) + lambda P (w)]Wherein L (w) is a loss function; λ is a regularization factor; argmin () is the value of the argument that minimizes the parenthetical inner formula; p (w) is a penalty term; regularization is divided into L1 regression and L2 regression, and p (w) | | w | | survival in L1 regression1L2 regression
The ridge regression used the L2 regression, and the Lasso regression used the L1 regression.
10. The method of claim 2, wherein the step S9 of generating and processing the first signal test data T1 comprises the steps of:
(1) generating data by adopting a Matlab writing program;
(2) for the MPSK/MQAM modulation signal, the set modulation orders M comprise 2, 4, 8, 16, 32 and 64; for Gaussian pulse signals, a modulation order does not need to be set;
(3) setting an estimated SNR range according to the interested research range;
(4) symbol up-sampling frequency NssSetting the frequency to be 2 times or more of the signal frequency;
(5) setting the number of one-time transmission signals NsymThe number of samples, numbits, corresponding to each SNR;
(6) setting the order orderRcos of the formed filter;
(7) carrying out normalization processing on the generated data;
the step S10 is to extract and process the time domain feature of the second signal test data T2, so as to obtain second signal test feature data FT1, specifically, the following steps are adopted for extraction and processing:
a. for a gaussian pulse signal, the extracted time domain features include: a maximum, a minimum, a mean, a peak-to-peak, a rectified mean, a variance, a standard deviation, a kurtosis, a root mean square, a form factor, a peak factor, a kurtosis factor, a pulse factor, and a margin factor;
b. for MPSK/MQAM modulated signals, the signals are divided into real parts and imaginary parts, and time domain characteristics which are respectively extracted according to real part data and data which need to be complemented comprise: a maximum, a minimum, a mean, a peak-to-peak, a rectified mean, a variance, a standard deviation, a kurtosis, a root mean square, a form factor, a peak factor, a kurtosis factor, a pulse factor, and a margin factor;
c. b, performing normalization processing on the time domain features extracted in the step a and the step b to obtain final second signal test feature data FT 1;
step S11, extracting the features of the second signal test data T2 obtained in step S9 by using the pre-training signal feature extraction model obtained in step S5, so as to obtain a second signal test data feature FT2, specifically, inputting the second signal test data T2 to the pre-training signal feature extraction model, and obtaining the features extracted by the second full-link sublayer in the pre-training signal feature extraction model, and using the features as the second signal test data features;
step S12, fusing the second signal test feature data FT1 obtained in step S10 and the second signal test data feature FT2 obtained in step S11 to obtain a second combined feature, specifically, fusing the second signal test feature data FT1 obtained in step S10 and the second signal test data feature FT2 obtained in step S11 in a manner of FT1+ FT2 to obtain a second combined feature;
the performance evaluation of the ridge regression signal-to-noise ratio estimation model described in step S13 is specifically performed by using Bias, normalized MEAN, variance VAR, normalized Bias NBias, and normalized variance NMSE as evaluation indexes to evaluate the ridge regression signal-to-noise ratio estimation model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110239443.6A CN112906601B (en) | 2021-03-04 | 2021-03-04 | Signal-to-noise ratio estimation method for multiple signals |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110239443.6A CN112906601B (en) | 2021-03-04 | 2021-03-04 | Signal-to-noise ratio estimation method for multiple signals |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112906601A CN112906601A (en) | 2021-06-04 |
| CN112906601B true CN112906601B (en) | 2022-06-03 |
Family
ID=76107009
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110239443.6A Active CN112906601B (en) | 2021-03-04 | 2021-03-04 | Signal-to-noise ratio estimation method for multiple signals |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112906601B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114938248B (en) * | 2022-07-26 | 2022-11-01 | 中国海洋大学三亚海洋研究院 | Method for building and demodulating underwater wireless optical communication demodulation model |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110610458A (en) * | 2019-04-30 | 2019-12-24 | 北京联合大学 | Method and system for GAN image enhancement interactive processing based on ridge regression |
| WO2020245748A1 (en) * | 2019-06-03 | 2020-12-10 | Polyvalor, Limited Partnership | Methods and systems for assessing a phenotype of a biological tissue of a patient using raman spectroscopy |
-
2021
- 2021-03-04 CN CN202110239443.6A patent/CN112906601B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110610458A (en) * | 2019-04-30 | 2019-12-24 | 北京联合大学 | Method and system for GAN image enhancement interactive processing based on ridge regression |
| WO2020245748A1 (en) * | 2019-06-03 | 2020-12-10 | Polyvalor, Limited Partnership | Methods and systems for assessing a phenotype of a biological tissue of a patient using raman spectroscopy |
Non-Patent Citations (1)
| Title |
|---|
| 朱凌云等.噪声环境下起搏心电信号的压缩感知重构算法.《计算机工程与应用》.2017,(第18期), * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112906601A (en) | 2021-06-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109657604B (en) | Satellite amplitude-phase signal identification and demodulation method and device based on cyclic neural network | |
| CN107124381B (en) | Automatic identification method for digital communication signal modulation mode | |
| CN108764077B (en) | Digital signal modulation classification method based on convolutional neural network | |
| CN111783558A (en) | Satellite navigation interference signal type intelligent identification method and system | |
| CN108169708B (en) | Direct positioning method of modular neural network | |
| CN110532932B (en) | Method for identifying multi-component radar signal intra-pulse modulation mode | |
| CN113114599B (en) | Modulation identification method based on lightweight neural network | |
| CN112347871A (en) | Interference signal modulation identification method for communication carrier monitoring system | |
| CN112906601B (en) | Signal-to-noise ratio estimation method for multiple signals | |
| CN113723556B (en) | Modulation mode identification method based on entropy weighting-multi-mode domain antagonistic neural network | |
| Ali et al. | Algorithm for automatic recognition of PSK and QAM with unique classifier based on features and threshold levels | |
| CN115186712A (en) | Modulated signal identification method and system | |
| CN111612130A (en) | Frequency shift keying communication signal modulation mode identification method | |
| CN113076925B (en) | M-QAM signal modulation mode identification method based on CNN and ELM | |
| CN113095162B (en) | Spectrum sensing method based on semi-supervised deep learning | |
| CN116760491A (en) | Signal-to-noise ratio estimation method based on deep learning | |
| He et al. | Deep learning-based automatic modulation recognition algorithm in non-cooperative communication systems | |
| CN115563485A (en) | Low-complexity interference identification method based on deep learning | |
| Zhang et al. | A Pruning Neural Network for Automatic Modulation Classification | |
| Zhao et al. | A spatiotemporal multi-channel learning framework for PCMA signal detection and recognition algorithm | |
| Zhu et al. | Application of machine learning in wireless communications | |
| Fontes et al. | Novel automatic modulation classification using correntropy coefficient | |
| CN112202510B (en) | Signal detection method based on regularization stream neural network | |
| CN116361728B (en) | Civil aircraft system level abnormal precursor identification method based on real-time flight data | |
| CN114301499B (en) | Intelligent frequency hopping signal receiving method based on CNN-GRU |
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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20231012 Address after: Room 415, 4th Floor, Building A1, No. 2 Courtyard of Shaoyaoju, Chaoyang District, Beijing, 100020 Patentee after: Guoke Blue Shield (Beijing) Technology Co.,Ltd. Address before: No. 932, Lushan South Road, Yuelu District, Changsha City, Hunan Province Patentee before: CENTRAL SOUTH University |