CN115913849A - Electromagnetic signal identification method based on one-dimensional complex value residual error network - Google Patents
Electromagnetic signal identification method based on one-dimensional complex value residual error network Download PDFInfo
- Publication number
- CN115913849A CN115913849A CN202211350504.7A CN202211350504A CN115913849A CN 115913849 A CN115913849 A CN 115913849A CN 202211350504 A CN202211350504 A CN 202211350504A CN 115913849 A CN115913849 A CN 115913849A
- Authority
- CN
- China
- Prior art keywords
- complex
- complex value
- signal
- residual error
- resnet
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000012360 testing method Methods 0.000 claims abstract description 20
- 238000013528 artificial neural network Methods 0.000 claims abstract description 16
- 238000010606 normalization Methods 0.000 claims abstract description 14
- 238000012549 training Methods 0.000 claims abstract description 12
- 238000004891 communication Methods 0.000 claims abstract description 11
- 238000012952 Resampling Methods 0.000 claims abstract description 5
- 230000008034 disappearance Effects 0.000 claims abstract description 5
- 238000001914 filtration Methods 0.000 claims abstract description 5
- 238000012545 processing Methods 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 16
- 238000013507 mapping Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000013499 data model Methods 0.000 claims description 2
- 238000004880 explosion Methods 0.000 claims description 2
- 238000011160 research Methods 0.000 abstract description 2
- 238000013135 deep learning Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000005562 fading Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000012706 support-vector machine 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
- Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
Abstract
The invention provides an electromagnetic signal identification method based on a one-dimensional complex value residual error network, which comprises the steps of converting a communication signal received by a receiver into signal data in an int16 format, carrying out Fourier transform, drawing a spectrogram of the signal, estimating the carrier frequency of the signal, resampling the signal, carrying out power normalization after intermediate frequency filtering, and training a data set subjected to power normalization processing through the one-dimensional complex value residual error network, thereby calculating the accuracy of a model on a test set. The method solves the problem of signal feature loss caused by training of the neural network by real number representation of the communication signal in the existing research, and simultaneously, the residual error network can avoid the problem of gradient disappearance, so that a deeper network can be designed to extract the original features of the communication signal, and higher accuracy of modulation recognition is realized.
Description
Technical Field
The invention relates to the field of communication, in particular to a modulation identification method of a wireless communication signal, which is suitable for identifying a modulation mode of an electromagnetic signal by adopting a deep learning technology.
Background
Automatic Modulation Recognition (AMR) of electromagnetic signals is used as an intermediate process between signal detection and signal demodulation, provides signal Modulation information, and plays a key role in civil and military applications such as cognitive radio, signal Recognition, threat assessment, spectrum monitoring and the like.
Conventional electromagnetic signal identification methods can be divided into two categories: likelihood-based methods and feature-based methods. Likelihood-based methods generally achieve higher accuracy and minimize the probability of error, but such methods suffer from high delay classification or require full a priori knowledge, and are computationally expensive. The traditional method based on the characteristics mostly adopts classifiers such as decision trees, support vector machines and the like, and the classification accuracy is low.
In recent years, artificial intelligence methods represented by deep learning have made remarkable progress in electromagnetic signal recognition by using the learning ability of a network without extracting expert features. However, most of the current deep learning methods use real numbers of signals to represent a training neural network, and use less real-imaginary part associated information of complex numbers, thereby limiting further improvement of the electromagnetic signal identification accuracy.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides an electromagnetic signal identification method based on a one-dimensional complex value residual error network. Aiming at the defects of the prior art, the invention aims to provide a communication signal modulation identification method based on a one-dimensional complex residual network (ResNet). The problem of signal characteristic loss caused by training of a neural network by using real number representation of communication signals in the existing research is solved. Meanwhile, the residual error network can avoid the problem of gradient disappearance, so that a deeper network can be designed to extract the original characteristics of the communication signals, and higher accuracy of modulation identification is realized.
The technical scheme adopted by the invention for solving the technical problem comprises the following steps:
step 1: converting the communication signal received by the receiver into signal data in an int16 format;
step 2: carrying out Fourier transform on the signal data in the step 1, drawing a spectrogram of the signal, and estimating the carrier frequency of the signal;
and 3, step 3: resampling the signal data generated in step 1 with four times of carrier frequency;
and 4, step 4: performing intermediate frequency filtering on the signal obtained after the resampling in the step 3 according to the carrier frequency estimated in the step 2;
and 5: carrying out power normalization processing on the signals subjected to intermediate frequency filtering in the step 4;
step 6: dividing a data set subjected to power normalization processing into a training set and a test set, designing a one-dimensional complex value residual error network, training by using the one-dimensional complex value residual error network, and then sending the test set into a trained deep complex value attention machine neural network for testing to obtain a corresponding digital signal modulation mode;
and 7: model testing
And inputting the divided test set into the trained model, judging the output category of the model in the label category of each datum, and calculating the accuracy of the model on the test set.
The one-dimensional complex residual error network in the step 6 comprises 6 submodules, the 6 submodules are respectively a first one-dimensional complex convolution module, a first C-ResNet Block module, a second C-ResNet Block module, a third C-ResNet Block module, a fourth C-ResNet Block module and a full connection layer module, and the 6 submodules are sequentially connected in series.
The first C-ResNet Block module comprises 3 complex value ResNet building blocks, the second C-ResNet Block module comprises 4 complex value ResNet building blocks, the third C-ResNet Block module comprises 6 complex value ResNet building blocks, and the fourth C-ResNet Block module comprises 3 complex value ResNet building blocks.
The complex value ResNet building block comprises a first one-dimensional complex value convolution layer, a first one-dimensional complex Batchnorm layer, a first complex value ReLu activation function, a second one-dimensional complex value convolution layer, a second one-dimensional complex value Batchnorm layer, a first complex value downsample and a second complex value ReLU activation function, and the complex value convolution layer and the activation function are connected in series.
In the complex value convolution layer, in order to internally simulate a complex value algorithm by using a real value algorithm, a complex value convolution kernel weight matrix W = A + iB and a complex value vector s = x + iy are defined, which respectively represent I/Q signals, and the complex vector is convoluted by the obtained convolution kernel:
wherein, A and B are respectively the real part and the imaginary part of the weight matrix of the complex convolution kernel, and x and y are respectively the real part and the imaginary part of the complex vector;
then, the real and imaginary parts of the convolution result are obtained as follows:
each convolution layer is followed by a complex batch normalization layer, batch normalization is an important technology for optimizing a data model, and the real-value complex batch normalization layer is expressed as follows:
wherein,
and V is a Batch data mean and variance, respectively;γ and β are trainable parameters, σ is a very small value that prevents the denominator from being zero;
however for a one-dimensional complex-valued residual network:
a learnable shift parameter β and a scale parameter γ are set for complex batch normalization, the scale parameter γ given by:
shift parameter beta and shift parameter gamma in scale parameter gamma ri Will be initialized to zero and gamma in the scaling parameter gamma rr And gamma ii Is initialized to
In step 6, before the one-dimensional complex residual error network is trained, weight initialization is performed, correct initialization of the one-dimensional complex residual error network is crucial to reducing the risk of gradient disappearance or explosion, and the complex weight is expressed as:
W＝Real{W}+iImag{W}
when W is symmetrically distributed near 0, the variance of W is estimated from the parameter sigma of Rayleigh distribution, the weight amplitude is initialized according to Rayleigh distribution, the expected parameter is 0, and the variance parameter is 2 sigma 2 The parameter σ will be set differently according to different neural network architectures.
In the one-dimensional complex value residual error network, deeper neural networks can extract more distinguishing features for automatic modulation classification, but as the neural networks deepen, the training precision gradually saturates and deteriorates; the ResNet network can effectively solve the gradient disappearance problem through a unique residual error structure, and therefore the ResNet network is designed into a deeper network. The ResNet building block contains two mappings, a residual mapping and an identity mapping, where the residual mapping is described as:
F(z)＝H(z)-z
the expected bottom layer mapping of the ResNet block is shown, and under the structure, the network only needs to learn the difference between input and output, so that the learning difficulty is simplified, and the problem that deep gradient disappears is solved.
The one-dimensional complex residual error network adopts C-ResNet18 or C-ResNet34, and the structure is shown in FIG. 1 (a) and FIG. 1 (b).
The loss function of the one-dimensional complex value residual error network adopts a cross entropy loss function, the optimization function selects Adam, the initial learning rate lr =0.001, the learning rate modulation strategy adopts equal interval adjustment learning rate SteplR, the adjustment interval is 50Epochs, and the adjustment multiple is 0.1.
The invention has the beneficial effects that: first, the communication signal is represented by complex numbers, the real and imaginary parts of which are interdependent under any phase change caused by the time-shift effect. The real-valued network model will consider the real and imaginary parts of the signal to be independent, while the complex-valued neural network model will consider the correlation between the real and imaginary parts of the signal. Second, the amplitude and phase are important to learning the target, so it makes sense to use a complex-valued neural network model. The complex signal represents a single-variable complex parameter rather than two-variable real parameters, so that a solution space with less freedom degrees is provided.
Drawings
Fig. 1 is a one-dimensional complex residual error network structure proposed by the present invention, fig. 1 (a) is a C-ResNet18 network structure, and fig. 1 (b) is a C-ResNet34 network structure.
Fig. 2 is a graph showing the effect of comparing a one-dimensional complex residual network with a corresponding real-valued network based on a radiometl 2016.10b data set.
Detailed Description
The invention is further illustrated by the following examples in conjunction with the drawings.
As shown in fig. 1, a method for identifying electromagnetic signals based on a one-dimensional complex residual error network. The experiment of the present invention used the baseline open-source data set rml2016.10b. The signal-to-noise ratio of the three data sets ranges from dB to 18dB, the signals are sampled according to 4 samples/symbol during signal collection, each data sample is an IQ sequence with 128 sampling points, a large number of real voice signals are used in the generation process, and a dynamic channel model in GNU Radio is used for simulating various channel effects such as Carrier Frequency Offset (CFO), phase offset, white Gaussian noise (Gaussian white noise), fading, multipath and the like.
Step 1: the published standard modulated signal data set rml2016.10b data set is sampled.
Step 2: generating a modulation signal time-frequency spectrogram data set;
the rml2016.10b data set simulates various modulation signals in real life by using software Radio software GUN Radio, and comprises 11 modulation signals: 8PSK, BPSK, AM-DSB, QPSK, QAM16, QAM64, CPFSK, GFSK, 4PAM and WBFM; the signal-to-noise ratio coverage ranges from-20 dB to +18dB, the interval is 2dB, and the sampling length is 128. All signal data is collected from the signal at 4 samples per symbol and a sampling rate of 1M/s. The received signal is affected by radio channel imperfections. Each signal data is stored as a 2 x 128 matrix, with two rows of data corresponding to the in-phase and quadrature portions of the complex signal samples, respectively.
And 2, step: labeling data;
the data set contains 10 modulation modes, and each signal data is labeled with an SNR and a modulation mode.
And step 3: dividing a data set;
in order to obtain a training set and a test set, the invention adopts a random division mode under a certain modulation mode of a certain signal-to-noise ratio. That is, for each modulation type signal, 80% of the data of each signal-to-noise ratio value is randomly selected as a training set, and 20% is selected as a test set.
And 4, step 4: building a network model
A structure is built by adopting a Pythrch deep learning framework.
And 5: model training
The deep neural network loss function selects a cross entropy loss function, and the deep neural network optimization function selects Adam. The initial learning rate lr =0.001 was selected. The learning rate modulation strategy adopts equal interval adjustment learning rate StepLR, the adjustment interval is 50Epochs, and the adjustment multiple is 0.1.
When 150Epochs were trained, the test set accuracy reached the highest. Selecting a neural network parameter value file stored when the neural network parameter value file is 150Epochs, and reserving the neural network parameter value file as a deep neural network for use when the deep neural network is in a test mode.
And 7: model testing
Inputting the divided test set into the trained model, judging the output category of the model to the label category of each datum, and calculating the accuracy of the model on the test set.
To evaluate the performance of the proposed one-dimensional replica residual network, the modulation classification accuracy of C-ResNet34, C-ResNet18, resNet34, and ResNet18 at different signal-to-noise ratios were compared, as shown in fig. 1. Under the condition of relatively high signal-to-noise ratio (0-18 db), the performance of the proposed complex value model is improved by 3-10% compared with that of a corresponding real value model. At relatively low signal-to-noise levels (-20-0 dB), the one-dimensional complex-valued model of the present invention yields 2-2.2dB gain over the corresponding real-valued network model at the same accuracy. The complex value model can better learn the time-frequency domain characteristics of the signal data, and the anti-noise capability of the modulation recognition network is improved.
Meanwhile, the C-ResNet34 is observed to show better performance than the C-ResNet18, which also shows that a deeper residual network can fit more complex function characteristics, so that the characteristic extraction capability is stronger, and the modulation identification accuracy is higher.
Claims (9)
1. An electromagnetic signal identification method based on a one-dimensional complex value residual error network is characterized by comprising the following steps:
step 1: converting the communication signal received by the receiver into signal data in an int16 format;
step 2: carrying out Fourier transform on the signal data in the step 1, drawing a spectrogram of the signal, and estimating the carrier frequency of the signal;
and step 3: resampling the signal data generated in the step 1 by using four times of carrier frequency;
and 4, step 4: performing intermediate frequency filtering on the signal obtained after the resampling in the step 3 according to the carrier frequency estimated in the step 2;
and 5: carrying out power normalization processing on the signals subjected to intermediate frequency filtering in the step 4;
step 6: dividing a data set subjected to power normalization processing into a training set and a test set, designing a one-dimensional complex value residual error network, training by using the one-dimensional complex value residual error network, and then sending the test set into a trained deep complex value attention machine neural network for testing to obtain a corresponding digital signal modulation mode;
and 7: model testing
And inputting the divided test set into the trained model, judging the output category of the model in the label category of each datum, and calculating the accuracy of the model on the test set.
2. The method of claim 1, wherein the method comprises:
the one-dimensional complex residual error network in the step 6 comprises 6 submodules, the 6 submodules are respectively a first one-dimensional complex convolution module, a first C-ResNet Block module, a second C-ResNet Block module, a third C-ResNet Block module, a fourth C-ResNet Block module and a full connection layer module, and the 6 submodules are sequentially connected in series.
3. The method of claim 2, wherein the method comprises:
the first C-ResNet Block module comprises 3 complex value ResNet building blocks, the second C-ResNet Block module comprises 4 complex value ResNet building blocks, the third C-ResNet Block module comprises 6 complex value ResNet building blocks, and the fourth C-ResNet Block module comprises 3 complex value ResNet building blocks.
4. The method of claim 3, wherein the method comprises:
the complex value ResNet building block comprises a first one-dimensional complex value convolution layer, a first one-dimensional complex Batchnorm layer, a first complex value ReLu activation function, a second one-dimensional complex value convolution layer, a second one-dimensional complex value Batchnorm layer, a first complex value downsample and a second complex value ReLU activation function, and the complex value convolution layer and the activation functions are connected in series.
5. The method of claim 4, wherein the method comprises:
in the complex value convolution layer, in order to internally simulate a complex value algorithm by using a real value algorithm, a complex value convolution kernel weight matrix W = A + iB and a complex value vector s = x + iy are defined, which respectively represent I/Q signals, and the complex vector is convoluted by the obtained convolution kernel:
wherein, A and B are respectively the real part and the imaginary part of the weight matrix of the complex convolution kernel, and x and y are respectively the real part and the imaginary part of the complex vector;
then, the real and imaginary parts of the convolution result are obtained as follows:
each convolution layer is followed by a complex batch normalization layer, batch normalization is an important technology for optimizing a data model, and the real-value complex batch normalization layer is expressed as follows:
wherein,
and V is a Batch data mean and variance, respectively; γ and β are trainable parameters, and e is a very small value that prevents the denominator from being zero;
however for a one-dimensional complex-valued residual network:
setting a learnable shift parameter β and a scale parameter γ for complex batch normalization, the scale parameter γ given by:
shift parameter beta and shift parameter gamma in scale parameter gamma ri Will be initialized to zero and gamma in the scaling parameter gamma rr And gamma ii Is initialized to
6. The method of claim 1, wherein the method comprises:
in step 6, before the one-dimensional complex residual error network is trained, weight initialization is performed, correct initialization of the one-dimensional complex residual error network is crucial to reducing the risk of gradient disappearance or explosion, and the complex weight is expressed as:
W＝Real{W}+iImag{W}
when W is symmetrically distributed near 0, the variance of W is estimated from the parameter sigma of Rayleigh distribution, the weight amplitude is initialized according to Rayleigh distribution, the expected parameter is 0, and the variance parameter is 2 sigma 2 The parameter σ will be set differently according to different neural network architectures.
7. The method for identifying electromagnetic signals based on one-dimensional complex residual error network according to any one of claims 3-6, wherein:
in the one-dimensional complex value residual error network, a ResNet building block comprises two mappings, namely residual error mapping and identity mapping, wherein the residual error mapping is described as follows:
F(z)＝H(z)-z
representing the expected underlying mapping of the ResNet block, under this structure the network only needs to learn the difference of inputs and outputs.
8. The method of claim 1, wherein the method comprises: the one-dimensional complex residual network adopts C-ResNet18 or C-ResNet34.
9. The method of claim 1, wherein the method comprises:
the loss function of the one-dimensional complex value residual error network adopts a cross entropy loss function, the optimization function selects Adam, the initial learning rate lr =0.001, the learning rate modulation strategy adopts equal interval adjustment learning rate SteplR, the adjustment interval is 50Epochs, and the adjustment multiple is 0.1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211350504.7A CN115913849A (en) | 2022-10-31 | 2022-10-31 | Electromagnetic signal identification method based on one-dimensional complex value residual error network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211350504.7A CN115913849A (en) | 2022-10-31 | 2022-10-31 | Electromagnetic signal identification method based on one-dimensional complex value residual error network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115913849A true CN115913849A (en) | 2023-04-04 |
Family
ID=86473661
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211350504.7A Pending CN115913849A (en) | 2022-10-31 | 2022-10-31 | Electromagnetic signal identification method based on one-dimensional complex value residual error network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115913849A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116992255A (en) * | 2023-07-13 | 2023-11-03 | 华北电力大学 | Screening method and system for transient voltage stability characteristic quantity and electronic equipment |
-
2022
- 2022-10-31 CN CN202211350504.7A patent/CN115913849A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116992255A (en) * | 2023-07-13 | 2023-11-03 | 华北电力大学 | Screening method and system for transient voltage stability characteristic quantity and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110855591B (en) | QAM and PSK signal intra-class modulation classification method based on convolutional neural network structure | |
| CN110163282B (en) | Modulation mode identification method based on deep learning | |
| CN109657604B (en) | Satellite amplitude-phase signal identification and demodulation method and device based on cyclic neural network | |
| CN110300078B (en) | Modulated signal identification method based on course learning | |
| CN110598677B (en) | Space-time multi-channel deep learning system for automatic modulation recognition | |
| CN108234370A (en) | Modulation mode of communication signal recognition methods based on convolutional neural networks | |
| Zhang et al. | A data preprocessing method for automatic modulation classification based on CNN | |
| CN109672639B (en) | Signal demodulation method based on machine learning | |
| CN113014524B (en) | Digital signal modulation identification method based on deep learning | |
| CN112464837A (en) | Shallow sea underwater acoustic communication signal modulation identification method and system based on small data samples | |
| CN112733811B (en) | Method for identifying underwater sound signal modulation modes based on improved dense neural network | |
| CN112347871B (en) | Interference signal modulation identification method for communication carrier monitoring system | |
| CN109120563B (en) | Modulation recognition method based on neural network integration | |
| CN110417694A (en) | A kind of modulation mode of communication signal recognition methods | |
| CN112242969A (en) | Novel single-bit OFDM receiver based on model-driven deep learning | |
| CN111131108B (en) | Non-cooperative underwater sound OFDM subcarrier modulation mode identification method | |
| CN111431825B (en) | Signal automatic classification and identification method based on deep multi-flow neural network | |
| CN114422311A (en) | Signal modulation identification method and system combining deep neural network and expert prior characteristics | |
| CN115913849A (en) | Electromagnetic signal identification method based on one-dimensional complex value residual error network | |
| Juan-ping et al. | Automatic modulation recognition of digital communication signals | |
| CN117614467B (en) | Underwater sound signal intelligent receiving method based on noise reduction neural network | |
| CN114595729A (en) | Communication signal modulation identification method based on residual error neural network and meta-learning fusion | |
| CN113902095A (en) | Automatic modulation identification method, device and system for wireless communication | |
| Wang et al. | Automatic modulation classification based on CNN, LSTM and attention mechanism | |
| Kalade et al. | Using sequence to sequence learning for digital bpsk and qpsk demodulation |
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 |