CN117614467A - Underwater sound signal intelligent receiving method based on noise reduction neural network - Google Patents
Underwater sound signal intelligent receiving method based on noise reduction neural network Download PDFInfo
- Publication number
- CN117614467A CN117614467A CN202410063273.4A CN202410063273A CN117614467A CN 117614467 A CN117614467 A CN 117614467A CN 202410063273 A CN202410063273 A CN 202410063273A CN 117614467 A CN117614467 A CN 117614467A
- Authority
- CN
- China
- Prior art keywords
- signal
- noise
- representing
- model
- neural network
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 61
- 230000009467 reduction Effects 0.000 title claims abstract description 44
- 230000005236 sound signal Effects 0.000 title claims abstract description 38
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 33
- 238000009792 diffusion process Methods 0.000 claims abstract description 47
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000003062 neural network model Methods 0.000 claims abstract description 18
- 238000000605 extraction Methods 0.000 claims abstract description 11
- 238000007781 pre-processing Methods 0.000 claims abstract description 3
- 230000006870 function Effects 0.000 claims description 31
- 238000009826 distribution Methods 0.000 claims description 21
- 238000004364 calculation method Methods 0.000 claims description 20
- 238000011176 pooling Methods 0.000 claims description 15
- 238000005070 sampling Methods 0.000 claims description 15
- 238000010606 normalization Methods 0.000 claims description 14
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000005457 optimization Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 9
- 108010076504 Protein Sorting Signals Proteins 0.000 claims description 6
- 238000013527 convolutional neural network Methods 0.000 claims description 5
- 230000005540 biological transmission Effects 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 4
- 230000002708 enhancing effect Effects 0.000 claims description 4
- LJROKJGQSPMTKB-UHFFFAOYSA-N 4-[(4-hydroxyphenyl)-pyridin-2-ylmethyl]phenol Chemical compound C1=CC(O)=CC=C1C(C=1N=CC=CC=1)C1=CC=C(O)C=C1 LJROKJGQSPMTKB-UHFFFAOYSA-N 0.000 claims description 3
- 230000007787 long-term memory Effects 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 230000007246 mechanism Effects 0.000 claims description 3
- 230000015654 memory Effects 0.000 claims description 3
- 230000008447 perception Effects 0.000 claims description 3
- 230000006403 short-term memory Effects 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000000844 transformation Methods 0.000 claims description 3
- 238000009827 uniform distribution Methods 0.000 claims description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 abstract 3
- 238000004891 communication Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B1/00—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission
- H04B1/06—Receivers
- H04B1/10—Means associated with receiver for limiting or suppressing noise or interference
- H04B1/1027—Means associated with receiver for limiting or suppressing noise or interference assessing signal quality or detecting noise/interference for the received signal
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B11/00—Transmission systems employing sonic, ultrasonic or infrasonic waves
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B13/00—Transmission systems characterised by the medium used for transmission, not provided for in groups H04B3/00 - H04B11/00
- H04B13/02—Transmission systems in which the medium consists of the earth or a large mass of water thereon, e.g. earth telegraphy
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L27/00—Modulated-carrier systems
- H04L27/0008—Modulated-carrier systems arrangements for allowing a transmitter or receiver to use more than one type of modulation
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L27/00—Modulated-carrier systems
- H04L27/0012—Modulated-carrier systems arrangements for identifying the type of modulation
-
- 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
Abstract
The invention relates to the technical field of underwater acoustic signal noise reduction, in particular to an intelligent underwater acoustic signal receiving method based on a noise reduction neural network. The method specifically comprises the steps of generating a water sound signal sample for training, preprocessing the sample, respectively establishing a noise reduction diffusion probability model and a mixed depth neural network model for feature extraction, acquiring a representation with more robustness of the water sound signal by using the noise reduction diffusion probability model, improving the signal to noise ratio, performing two-stage training by using the mixed depth neural network model, respectively completing automatic recognition of a modulation mode and signal self-adaptive demodulation tasks, and realizing intelligent receiving of the water sound signal. The technical problems that the error rate is high, the implementation is complex and only a single signal can be aimed at in the current signal receiving method are solved.
Description
Technical Field
The invention relates to the technical field of underwater acoustic signal noise reduction, in particular to an intelligent underwater acoustic signal receiving method based on a noise reduction neural network.
Background
The underwater acoustic communication is widely applied to the fields of marine science research, underwater detection, underwater robots, underwater military attack and defense and the like. However, due to the complexity and dynamic characteristics of the underwater acoustic channel, the signal is easily affected by interference, attenuation and distortion in the transmission process, the receiving end is difficult to better recover and reconstruct the received signal, in addition, in the non-cooperative communication scene such as electronic countermeasure, the receiving end always receives a completely unknown signal, which greatly increases the difficulty of signal demodulation. Therefore, how to stably and efficiently receive and flexibly and intelligently demodulate the underwater acoustic signal is a hot spot of current research.
The noise reduction diffusion probability model is a probability-based machine learning method, and mainly comprises an encoder and a decoder. In the encoder stage, the model begins with an initial distribution of data, and each step adds noise to the data, gradually converting the data to pure noise, through successive steps. In the decoder stage, the model predicts the noise distribution in the feature map generated by the encoder, and denoising is performed step by step. The model can finally have better data modeling capability, and can convert the input original data into characteristic representations with less noise distribution.
The underwater acoustic signal receiving method based on deep learning can effectively extract the time domain characteristics of signals, directly restore the signals to original symbols in a classification mode, simplify the complex design flow of the traditional signal receiver, reduce the links needing to participate in design artificially, and reduce the implementation difficulty of the underwater acoustic signal receiver. However, the existing method greatly reduces the accuracy of receiving under the condition of low signal-to-noise ratio, and can only singly extract the phase or frequency characteristics of the signals, namely, can only be effective for one or a few modulation modes, and cannot meet the complex special scenes of non-cooperative communication, encrypted communication and the like, which need to flexibly receive the signals.
Therefore, aiming at the above situations, a method with high accuracy and high flexibility is urgently needed to realize the high-efficiency intelligent receiving of the underwater sound signals.
Disclosure of Invention
The invention aims to provide an underwater sound signal intelligent receiving method based on a noise reduction neural network, which is characterized in that a noise reduction diffusion probability model and a mixed depth neural network model are respectively established for feature extraction by carrying out feature optimization processing on a signal sample, the noise reduction diffusion probability model is utilized to obtain the representation of the underwater sound signal with more robustness, the signal to noise ratio is improved, the mixed depth neural network model is utilized to carry out two-stage training, the automatic recognition of a modulation mode and the adaptive demodulation task of a signal are respectively completed, the high-efficiency intelligent receiving of the signal of an unknown modulation mode can be realized, and the accuracy and the flexibility are high.
In order to achieve the above purpose, the underwater sound signal intelligent receiving method based on the noise reduction neural network provided by the invention comprises the following steps:
s1, generating underwater sound signal samples containing multiple modulation modes for training, preprocessing the samples, and respectively optimizing time domain features, interpolating signals and normalizing periods;
s2, establishing an improved noise reduction diffusion probability model (Denoising Diffusion Probabilistic Models, DDPM), training the model by using the signals processed by the S1, and gradually adding Gaussian noise to the signals in the stage of establishing a forward noise reduction diffusion probability model; in the stage of establishing a reverse noise reduction diffusion probability model, predicting Gaussian noise distribution of signal samples, establishing a U-Net neural network, training, extracting signal characteristics, and reducing noise of signals;
s3, establishing a mixed depth neural network model formed by a convolution layer, a circulation layer and an attention layer, extracting signal characteristics, and taking a modulation mode of signals as a label based on the signals processed by the S1 and the S2; training a hybrid neural network model by using the modulation tag data set, and identifying a modulation mode of the underwater sound signal;
s4, constructing a data set based on the signals processed by the S1 and the S2, wherein each sample contains sampling points with the number corresponding to a single symbol, and the model classifies the original symbol of the signal as a label; the signals are demodulated by training a hybrid neural network model using the raw tag dataset.
Preferably, S1 generates underwater sound signal samples containing multiple modulation modes for training, performs feature optimization processing on the samples, interpolates each symbol sample of the generated underwater sound signal by adopting a radial basis function interpolation method, and performs period normalization on the signal.
Preferably, the step S1 specifically includes:
s1.1 generating underwater sound signals and performing feature optimization processing
Generating a modulated signal waveform by utilizing MATLAB, mapping sampling points representing single symbols to two orthogonal paths and splicing to obtain a signal waveform with typical characteristics of amplitude and phase easy to distinguish; the baseband modulation signal is as follows:
；
wherein,representing baseband modulation signal, ">Is the real part of the amplitude of the signal on the in-phase component, < >>Is the imaginary part of the amplitude of the signal on the quadrature component, < >>Representation->Real part of->Representation->Is the imaginary part of (2);
the baseband modulated signal will be multiplied by a carrier wave of fixed frequency for transmission in the underwater wireless channel and received and processed by the receiver; the final signal sequence after feature optimization is as follows:
；
wherein,representing the processed amplitude-phase characteristics, +.>Representing a carrier frequency;
s1.2, adopting a radial basis function interpolation method to interpolate the underwater sound signal
For signal waveforms with single symbol and smaller number of sampling points, selecting original sampling points as central points of radial basis functions, and determining Gaussian width parametersFor each center point, calculating a weight coefficient using a least square method; constructing an approximation function for signal interpolation by using the determined center point, gaussian width parameter and weight coefficient; the approximation function can be expressed as:
；
wherein,is a constructed approximation function, +.>Is the number of center points, +.>Is the weight coefficient corresponding to each center point, < ->Is the position of each center point, +.>The width parameter is the width parameter of the Gaussian function, and the width of the basis function is controlled;
s1.3 cycle normalization of signals
The signal is subjected to min-max normalization in each carrier period of the signal waveform, and the specific formula is as follows:
；
wherein,is the original signal amplitude, +.>Is the minimum amplitude in the carrier period, +.>Is the maximum amplitude within the carrier period.
Preferably, the step S2 specifically includes:
s2.1, establishing a forward noise reduction diffusion probability model
In the forward process, the input original signal is gradually superimposed with Gaussian noise to form a Markov chain for a determined data distributionThe procedure for each step is as follows:
；
；
wherein,indicate by->Obtain->Is>Finger adding noise->Post-time signal dataIs the data of the previous moment, +.>Representing a normal distribution, < >>Variance at each noise addition, +.>Representing an identity matrix for determining the speed and amplitude of diffusion,/->Represents time 1 to +.>A data set obtained at each moment; the formula is expressed by ∈>Obtain->Meeting the requirement of->For mean value->As the Gaussian distribution of the variance, the original signal gradually becomes pure noise after being subjected to noise addition;
s2.2 establishing a reverse noise reduction diffusion probability model
The original noise-free signal is predicted based on the noise-containing signal generated by the forward process, and the process of each step is as follows:
；
；
wherein,representing noise plus->Post-secondary signal data,/>Signal data representing non-noisy, +.>Indicates the number of times of noise addition, +.>Representing an approximation model of the back diffusion process,/->Representing the generation of noise->The probability of the signal data of the next time,represented by->Obtain->Is>Represented by->Subject to +.>To->Values of the uniform distribution;
to calculate each step in the reversal processAccording to the known->And->The following posterior diffusion probability calculation formula is obtained:
；
wherein,finger adding noise->Post-secondary signal data,/>Signal data without noise +.>Is a fixed value between 0 and 1 determined at each noise addition,/for>Representation unitA matrix for determining the speed and amplitude of diffusion,indicating pass->Sub-noise processed signal->And untreated original signal->Difference estimation between->Representing posterior diffusion conditional probability,/-, for>Represents a normal distribution;
based on a Bayes formula, the posterior diffusion probability is calculated by adopting the following method:
；
wherein,representing noise plus->Post-secondary signal data,/>Signal data representing the non-noisy signal,representing posterior diffusion conditional probabilities;
s2.3 establishing noise in the reverse Process prediction Signal for the U-Net neural network
The U-Net neural network comprises an encoder and decoder structure, wherein a signal waveform is firstly input into the encoder formed by five continuous convolution blocks, the convolution blocks are formed by two convolution layers and one pooling layer, and feature extraction and dimension reduction are carried out on the signal, wherein the convolution structure is a convolution kernel of 3x 3; the formula of the convolution layer is shown below:
；
wherein,representing the input signal>Representing convolution kernel +.>Is the output value of the convolution result,is the element value of the input signal,/>Is the weight of the convolution kernel;
the formula of the pooling layer is shown below:
；
wherein,representing the input signal>Is the output value of the pooling result, +.>Is the element value within the pooling window, +.>Is pooled stride,/->Is the size of the pooling window;
the feature map with reduced output size of the encoder is restored to the original size through five layers of continuous deconvolution blocks, wherein the deconvolution blocks comprise two convolution layers with the convolution kernel size of 3x3 and one deconvolution layer, and the deconvolution layer is realized by horizontally and vertically overturning the convolution kernel and then carrying out convolution operation with the input feature map.
Preferably, the step S3 specifically includes:
s3.1 establishing feature extraction module
The signal firstly passes through a characteristic extraction module consisting of three layers of convolutional neural networks, and the phase and frequency information of the signal contained in the sampling points are extracted and mapped into a characteristic vector with fixed dimension; the calculation is performed here using the one-dimensional convolution formula:
；
wherein,is the input signal sequence,/->Is a convolution filter, < >>Representing the result sequence of a one-dimensional convolution operation, +.>Is the index position in the output sequence, +.>Is a convolution kernel->Index position of->Element value representing the input signal,/->Weights representing the convolution kernel;
s3.2 building classification module
The feature vector is input to a classification module consisting of two stacked two-way long and short term memory neural networks, which can be expressed by the following formula:
；
；
；
；
；
；
in the above-mentioned formula(s),、/>、/>the outputs of the forget gate, the input gate and the output gate are respectively +.>Is an S-shaped growth curve>Representing forgetful door computationAs a result of (I)>Is a weight matrix of forgetting gates, +.>Is the implicit state of the last time step, +.>Is the input of the current time step,/-, for example>Is a new candidate cell state, +.>And->Weight matrix of input gate and candidate cell state, respectively, +.>And->Is an input gate and a bias term for candidate cell states; />Representing time step->Is->Representing time step->In a module using a cell state, an implied state, a forget gate, an input gate and an output gate to control information flow;
s3.3 building Global perception Module
A global attention sensing module based on a multi-head attention mechanism is added between two stacked two-way long-short-period memory neural networks and is used for enhancing the capturing capacity of a model on global information; the calculation method of the module is as follows:
；
；
；
wherein,、/>and->Are linear transformations of the original input vector multiplied by a specific coefficient, < >>Represents the firstSub-independent attention calculations, new output vector +.>，/>Connecting together a plurality of independent attention weighted sequence outputs, multiplying a weight to generate a final output +.>The method comprises the steps of carrying out a first treatment on the surface of the For each calculation, use +.>Function to generate an attentionOutput->，/>The calculation formula of (2) is as follows:
；
wherein,representative vector->Is>Element(s)>Is->Probability of individual category->The exponential term representing all classes is summed for normalization, ensuring that one probability distribution is output, since the sum of all probability values is 1;
in the final part of the neural network model, use is made ofOutputting signal modulation mode category by the function, calculating multi-classification cross entropy loss to evaluate the prediction effect of the model, and optimizing model parameters by adopting a self-adaptive moment estimation algorithm;
s3.4 training a model using pre-generated underwater acoustic signals
Generating signal waveforms by using various modulation modes such as BPSK, QPSK, 8PSK, 2FSK, 4FSK, 2ASK, 16QAM and the like, taking the signal waveforms with the fixed dimension of 1 x 1000 as samples, and taking the corresponding modulation modes as labels to generate a data set; on the basis of the generated data set, adding a plurality of Gaussian white noises with different signal to noise ratios respectively to form a noisy data set; training the models by using various data sets without noise and with noise respectively, and selecting the model with highest accuracy as a final modulation recognition model.
Preferably, the step S4 specifically includes:
generating signal waveforms by utilizing multiple modulation modes respectively, taking the signal waveforms with fixed dimensions as samples, taking corresponding original symbols as labels, and generating a data set; on the basis of the generated data set, adding a plurality of Gaussian white noises with different signal to noise ratios respectively to form a noisy data set; training the models by utilizing a plurality of non-noisy and noisy data sets respectively, and selecting the model with the highest accuracy as a final demodulation model; and respectively training the demodulation models offline according to each modulation mode, and automatically selecting the demodulation models according to the output result of the modulation recognition model in a test stage to realize intelligent underwater sound signal receiving.
The invention has the beneficial effects that:
according to the underwater sound signal intelligent receiving method based on the noise reduction neural network, firstly, feature optimization processing is carried out on the signal samples, each symbol sample of the generated underwater sound signal is interpolated by adopting a radial basis function interpolation method, and the signal is subjected to periodic normalization, so that time domain feature representation of the signal is enhanced, and the model can obtain more abundant feature details.
And secondly, noise of various scales of signals can be filtered by adopting a noise reduction diffusion probability model, the representation of the signals with more robustness is obtained, and the interference of the noise on the signals is reduced. And the phase and amplitude characteristics of the signals are extracted by using the mixed depth neural network model, the local characteristics of the signals are obtained by using the convolutional neural network, the time sequence characteristics are obtained by using the cyclic neural network, the modulation mode of the signals is identified and demodulated, the intelligent accurate receiving of the signals with unknown modulation mode can be realized, and the signal receiving efficiency is high, and the stability and the flexibility are strong.
Drawings
FIG. 1 is a schematic overall flow diagram of one embodiment of the present invention;
FIG. 2 is a schematic diagram of a noise reduction diffusion probability model structure of an embodiment of the present invention;
fig. 3 is a schematic diagram of a hybrid neural network classification model structure according to an embodiment of the invention.
Detailed Description
In order to make the technical means, the inventive features and the effects achieved by the present invention easy to understand, the technical solutions in the embodiments of the present invention will be further clearly and completely described below with reference to the drawings in the embodiments of the present invention. It will be apparent that the described embodiments are only some, but not all, embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention provides an underwater sound signal intelligent receiving method based on a noise reduction neural network, which is shown in figure 1 and comprises the following steps:
s1, generating underwater sound signal samples containing various modulation modes for training, carrying out feature optimization processing on the samples, interpolating each symbol sample of the generated underwater sound signal by adopting a radial basis function interpolation method, and carrying out period normalization on the signal, wherein the specific method comprises the following steps of:
s1.1 generating underwater sound signals and performing feature optimization processing
Generating a modulated signal waveform by utilizing MATLAB, mapping sampling points representing single symbols to two orthogonal paths and splicing to obtain a signal waveform with typical characteristics of amplitude and phase easy to distinguish; the baseband modulation signal is as follows:
；
wherein,representing baseband modulation signal, ">Is a signalReal part of amplitude on in-phase component, +.>Is the imaginary part of the amplitude of the signal on the quadrature component, < >>Representation->Real part of->Representation->Is the imaginary part of (2);
the baseband modulated signal will be multiplied by a carrier wave of fixed frequency for transmission in the underwater wireless channel and received and processed by the receiver; the final signal sequence after feature optimization is as follows:
；
wherein,representing the processed amplitude-phase characteristics, +.>Representing a carrier frequency;
s1.2, adopting a radial basis function interpolation method to interpolate the underwater sound signal
For signal waveforms with single symbol and smaller number of sampling points, selecting original sampling points as central points of radial basis functions, and determining Gaussian width parametersFor each center point, calculating a weight coefficient using a least square method; constructing an approximation function for signal interpolation by using the determined center point, gaussian width parameter and weight coefficient; the approximation function can be expressed as:
；
wherein,is a constructed approximation function, +.>Is the number of center points, +.>Is the weight coefficient corresponding to each center point, < ->Is the position of each center point, +.>The width parameter is the width parameter of the Gaussian function, and the width of the basis function is controlled;
s1.3 cycle normalization of signals
The signal is subjected to min-max normalization in each carrier period of the signal waveform, and the specific formula is as follows:
；
wherein,is the original signal amplitude, +.>Is the minimum amplitude in the carrier period, +.>Is the maximum amplitude within the carrier period.
S2, establishing an improved noise reduction diffusion probability model (Denoising Diffusion Probabilistic Models, DDPM), training the model by using the signals processed by the S1, and gradually adding Gaussian noise to the signals in the stage of establishing a forward noise reduction diffusion probability model; in the stage of establishing a reverse noise reduction diffusion probability model, predicting Gaussian noise distribution of a signal sample, establishing a U-Net neural network, training, extracting signal characteristics, and carrying out noise reduction on the signal, wherein the specific method comprises the following steps of:
s2.1, establishing a forward noise reduction diffusion probability model
In the forward process, the input original signal is gradually superimposed with Gaussian noise to form a Markov chain for a determined data distributionThe procedure for each step is as follows:
；
；
wherein,indicate by->Obtain->Is>Finger adding noise->Post-time signal dataIs the data of the previous moment, +.>Representing a normal distribution, < >>Is the variance at each noise addition, +.>Representing an identity matrix for determining the speed and amplitude of diffusion,/->Represents time 1 to +.>A data set obtained at each moment; the formula is expressed by ∈>Obtain->Meeting the requirement of->For mean value->As the Gaussian distribution of the variance, the original signal gradually becomes pure noise after being subjected to noise addition;
s2.2 establishing a reverse noise reduction diffusion probability model
The original noise-free signal is predicted based on the noise-containing signal generated by the forward process, and the process of each step is as follows:
；
；
wherein,representing noise plus->Post-secondary signal data,/>Signal data representing non-noisy, +.>Indicates the number of times of noise addition, +.>Representing an approximation model of the back diffusion process,/->Representing the generation of noise->The probability of the signal data of the next time,represented by->Obtain->Is>Represented by->Subject to +.>To->Values of the uniform distribution;
to calculate each step in the reversal processAccording to the known->And->The following posterior diffusion probability calculation formula is obtained:
；
wherein,finger adding noise->Post-secondary signal data,/>Signal data without noise +.>Is a fixed value between 0 and 1 determined at each noise addition,/for>Representing an identity matrix, for determining the speed and amplitude of diffusion,indicating pass->Sub-noise processed signal->And untreated original signal->Difference estimation between->Representing posterior diffusion conditional probability,/-, for>Represents a normal distribution;
based on a Bayes formula, the posterior diffusion probability is calculated by adopting the following method:
；
wherein,representing noise plus->Post-secondary signal data,/>Signal data representing the non-noisy signal,representing posterior diffusion conditional probabilities;
s2.3 establishing noise in the reverse Process prediction Signal for the U-Net neural network
The U-Net neural network comprises an encoder and decoder structure, wherein a signal waveform is firstly input into the encoder formed by five continuous convolution blocks, the convolution blocks are formed by two convolution layers and one pooling layer, and feature extraction and dimension reduction are carried out on the signal, wherein the convolution structure is a convolution kernel of 3x 3; the formula of the convolution layer is shown below:
；
wherein,representing the input signal>Representing convolution kernel +.>Is the output value of the convolution result,/>Is the element value of the input signal,/>Is the weight of the convolution kernel;
the formula of the pooling layer is shown below:
；
wherein,representing the input signal>Is the output value of the pooling result, +.>Is the element value within the pooling window, +.>Is pooled stride,/->Is the size of the pooling window;
the feature map with reduced output size of the encoder is restored to the original size through five layers of continuous deconvolution blocks, wherein the deconvolution blocks comprise two convolution layers with the convolution kernel size of 3x3 and one deconvolution layer, and the deconvolution layer is realized by horizontally and vertically overturning the convolution kernel and then carrying out convolution operation with the input feature map.
S3, establishing a mixed depth neural network model formed by a convolution layer, a circulation layer and an attention layer, extracting signal characteristics, and taking a modulation mode of signals as a label based on the signals processed by the S1 and the S2; training a hybrid neural network model by using a modulation tag data set, and identifying a modulation mode of an underwater sound signal, wherein the specific method comprises the following steps of:
s3.1 establishing feature extraction module
The signal firstly passes through a characteristic extraction module consisting of three layers of convolutional neural networks, and the phase and frequency information of the signal contained in the sampling points are extracted and mapped into a characteristic vector with fixed dimension; the calculation is performed here using the one-dimensional convolution formula:
；
wherein,is the input signal sequence,/->Is a convolution filter, < >>Representing the result sequence of a one-dimensional convolution operation, +.>Is the index position in the output sequence, +.>Is a convolution kernel->Index position of->Element value representing the input signal,/->Weights representing the convolution kernel;
s3.2 building classification module
The feature vector is input to a classification module consisting of two stacked two-way long and short term memory neural networks, which can be expressed by the following formula:
；
；
；
；
；
；
in the above-mentioned formula(s),、/>、/>the outputs of the forget gate, the input gate and the output gate are respectively +.>Is an S-shaped growth curve>Representing the result of a forgetting gate calculation +.>Is a weight matrix of forgetting gates, +.>Is the implicit state of the last time step, +.>Is the input of the current time step,/-, for example>Is a new candidate cell state, +.>And->Weight matrix of input gate and candidate cell state, respectively, +.>And->Is an input gate and a bias term for candidate cell states; />Representing time stepsIs->Representing time step->In a module using a cell state, an implied state, a forget gate, an input gate and an output gate to control information flow;
s3.3 building Global perception Module
A global attention sensing module based on a multi-head attention mechanism is added between two stacked two-way long-short-period memory neural networks and is used for enhancing the capturing capacity of a model on global information; the calculation method of the module is as follows:
；
；
；
wherein,、/>and->Are linear transformations of the original input vector multiplied by a specific coefficient, < >>Represents the firstSub-independent attention calculations, new output vector +.>，/>Connecting together a plurality of independent attention weighted sequence outputs, multiplying a weight to generate a final output +.>The method comprises the steps of carrying out a first treatment on the surface of the For each calculation, use +.>Function to generate an attention output +.>，/>The calculation formula of (2) is as follows:
；
wherein,representative vector->Is>Element(s)>Is->Probability of individual category->The exponential term representing all classes is summed for normalization, ensuring that one probability distribution is output, since the sum of all probability values is 1;
in the final part of the neural network model, use is made ofOutputting signal modulation mode category by the function, calculating multi-classification cross entropy loss to evaluate the prediction effect of the model, and optimizing model parameters by adopting a self-adaptive moment estimation algorithm;
s3.4 training a model using pre-generated underwater acoustic signals
Generating signal waveforms by using various modulation modes such as BPSK, QPSK, 8PSK, 2FSK, 4FSK, 2ASK, 16QAM and the like, taking the signal waveforms with the fixed dimension of 1 x 1000 as samples, and taking the corresponding modulation modes as labels to generate a data set; on the basis of the generated data set, adding a plurality of Gaussian white noises with different signal to noise ratios respectively to form a noisy data set; training the models by using various data sets without noise and with noise respectively, and selecting the model with highest accuracy as a final modulation recognition model.
S4, constructing a data set based on the signals processed by the S1 and the S2, wherein each sample contains sampling points with the number corresponding to a single symbol, and the model classifies the original symbol of the signal as a label; training a hybrid neural network model by using an original tag data set, and demodulating signals, wherein the specific method comprises the following steps of:
generating signal waveforms by utilizing multiple modulation modes respectively, taking the signal waveforms with fixed dimensions as samples, taking corresponding original symbols as labels, and generating a data set; on the basis of the generated data set, adding a plurality of Gaussian white noises with different signal to noise ratios respectively to form a noisy data set; training the models by utilizing a plurality of non-noisy and noisy data sets respectively, and selecting the model with the highest accuracy as a final demodulation model; and respectively training the demodulation models offline according to each modulation mode, and automatically selecting the demodulation models according to the output result of the modulation recognition model in a test stage to realize intelligent underwater sound signal receiving.
In summary, the method for intelligently receiving the underwater acoustic signal based on the noise reduction neural network carries out feature optimization processing on the signal samples, interpolates each symbol sample of the generated underwater acoustic signal by adopting a radial basis function interpolation method, and carries out period normalization on the signal, thereby enhancing the time domain feature representation of the signal and enabling the model to acquire richer feature details. And secondly, noise of various scales of signals can be filtered by adopting a noise reduction diffusion probability model, the representation of the signals with more robustness is obtained, and the interference of the noise on the signals is reduced. And the phase and amplitude characteristics of the signals are extracted by using the mixed depth neural network model, the local characteristics of the signals are obtained by using the convolutional neural network, the time sequence characteristics are obtained by using the cyclic neural network, the modulation mode of the signals is identified and demodulated, the intelligent accurate receiving of the signals with unknown modulation mode can be realized, and the signal receiving efficiency is high, and the stability and the flexibility are strong.
The above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be apparent to one skilled in the art that modifications may be made to the technical solutions described in the foregoing embodiments, or equivalents may be substituted for some of the technical features thereof; such modifications and substitutions do not depart from the spirit and scope of the corresponding technical solutions.
Claims (6)
1. The underwater sound signal intelligent receiving method based on the noise reduction neural network is characterized by comprising the following steps of:
s1, generating underwater sound signal samples containing multiple modulation modes for training, preprocessing the samples, and respectively optimizing time domain features, interpolating signals and normalizing periods;
s2, establishing an improved noise reduction diffusion probability model (Denoising Diffusion Probabilistic Models, DDPM), training the model by using the signals processed by the S1, and gradually adding Gaussian noise to the signals in the stage of establishing a forward noise reduction diffusion probability model; in the stage of establishing a reverse noise reduction diffusion probability model, predicting Gaussian noise distribution of signal samples, establishing a U-Net neural network, training, extracting signal characteristics, and reducing noise of signals;
s3, establishing a mixed depth neural network model formed by a convolution layer, a circulation layer and an attention layer, extracting signal characteristics, and taking a modulation mode of signals as a label based on the signals processed by the S1 and the S2; training a hybrid neural network model by using the modulation tag data set, and identifying a modulation mode of the underwater sound signal;
s4, constructing a data set based on the signals processed by the S1 and the S2, wherein each sample contains sampling points with the number corresponding to a single symbol, and the model classifies the original symbol of the signal as a label; the signals are demodulated by training a hybrid neural network model using the raw tag dataset.
2. The intelligent underwater sound signal receiving method based on the noise reduction neural network according to claim 1, wherein the step S1 is characterized in that the method comprises the steps of generating underwater sound signal samples containing a plurality of modulation modes for training, performing feature optimization processing on the samples, interpolating each symbol sample of the generated underwater sound signal by adopting a radial basis function interpolation method, and performing period normalization on the signal.
3. The method for intelligently receiving underwater sound signals based on the noise reduction neural network according to claim 2, wherein the step S1 is specifically:
s1.1 generating underwater sound signals and performing feature optimization processing
Generating a modulated signal waveform by utilizing MATLAB, mapping sampling points representing single symbols to two orthogonal paths and splicing to obtain a signal waveform with typical characteristics of amplitude and phase easy to distinguish; the baseband modulation signal is as follows:
；
wherein,representing baseband modulation signal, ">Is the real part of the amplitude of the signal on the in-phase component, < >>Is the imaginary part of the amplitude of the signal on the quadrature component, < >>Representation->Real part of->Representation->Is the imaginary part of (2);
the baseband modulated signal will be multiplied by a carrier wave of fixed frequency for transmission in the underwater wireless channel and received and processed by the receiver; the final signal sequence after feature optimization is as follows:
；
wherein,representing the treatedAmplitude-phase characteristics, +.>Representing a carrier frequency;
s1.2, adopting a radial basis function interpolation method to interpolate the underwater sound signal
For signal waveforms with single symbol and smaller number of sampling points, selecting original sampling points as central points of radial basis functions, and determining Gaussian width parametersFor each center point, calculating a weight coefficient using a least square method; constructing an approximation function for signal interpolation by using the determined center point, gaussian width parameter and weight coefficient; the approximation function can be expressed as:
；
wherein,is a constructed approximation function, +.>Is the number of center points, +.>Is the weight coefficient corresponding to each center point, < ->Is the position of each center point, +.>The width parameter is the width parameter of the Gaussian function, and the width of the basis function is controlled;
s1.3 cycle normalization of signals
The signal is subjected to min-max normalization in each carrier period of the signal waveform, and the specific formula is as follows:
；
wherein,is the original signal amplitude, +.>Is the minimum amplitude in the carrier period, +.>Is the maximum amplitude within the carrier period.
4. The method for intelligently receiving underwater sound signals based on the noise reduction neural network according to claim 1, wherein the step S2 is specifically:
s2.1, establishing a forward noise reduction diffusion probability model
In the forward process, the input original signal is gradually superimposed with Gaussian noise to form a Markov chain for a determined data distributionThe procedure for each step is as follows:
；
；
wherein,indicate by->Obtain->Is>Finger adding noise->Post-time signal dataIs the data of the previous moment, +.>Representing a normal distribution, < >>Is the variance at each noise addition, +.>Representing an identity matrix for determining the speed and amplitude of diffusion,/->Represents time 1 to +.>A data set obtained at each moment; the formula is expressed by ∈>Obtain->Meeting the requirement of->For mean value->As the Gaussian distribution of the variance, the original signal gradually becomes pure noise after being subjected to noise addition;
s2.2 establishing a reverse noise reduction diffusion probability model
The original noise-free signal is predicted based on the noise-containing signal generated by the forward process, and the process of each step is as follows:
；
；
wherein,representing noise plus->Post-secondary signal data,/>Signal data representing non-noisy, +.>Indicates the number of times of noise addition, +.>Representing an approximation model of the back diffusion process,/->Representing the generation of noise->Probability of secondary signal data,/>Represented by->Obtain->Is>Represented by->Subject to +.>To->Values of the uniform distribution;
to calculate each step in the reversal processAccording to the known->And->The following posterior diffusion probability calculation formula is obtained:
；
wherein,finger adding noise->Post-secondary signal data,/>Signal data without noise +.>Is a fixed value between 0 and 1 determined at each noise addition,/for>Representing an identity matrix for determining the speed and amplitude of diffusion,/->Indicating pass->Sub-noise processed signal->And untreated original signal->A difference estimate value between the two,representing posterior diffusion conditional probability,/-, for>Represents a normal distribution;
based on a Bayes formula, the posterior diffusion probability is calculated by adopting the following method:
；
wherein,representing noise plus->Post-secondary signal data,/>Signal data representing the non-noisy signal,representing posterior diffusion conditional probabilities;
s2.3 establishing noise in the reverse Process prediction Signal for the U-Net neural network
The U-Net neural network comprises an encoder and decoder structure, wherein a signal waveform is firstly input into the encoder formed by five continuous convolution blocks, the convolution blocks are formed by two convolution layers and one pooling layer, and feature extraction and dimension reduction are carried out on the signal, wherein the convolution structure is a convolution kernel of 3x 3; the formula of the convolution layer is shown below:
；
wherein,representing the input signal>Representing convolution kernel +.>Is the output value of the convolution result,/>Is the element value of the input signal,/>Is the weight of the convolution kernel;
the formula of the pooling layer is shown below:
；
wherein,representing the input signal>Is the output value of the pooling result, +.>Is the element value within the pooling window, +.>Is pooled stride,/->Is the size of the pooling window;
the feature map with reduced output size of the encoder is restored to the original size through five layers of continuous deconvolution blocks, wherein the deconvolution blocks comprise two convolution layers with the convolution kernel size of 3x3 and one deconvolution layer, and the deconvolution layer is realized by horizontally and vertically overturning the convolution kernel and then carrying out convolution operation with the input feature map.
5. The method for intelligently receiving underwater sound signals based on the noise reduction neural network according to claim 1, wherein the step S3 is specifically:
s3.1 establishing feature extraction module
The signal firstly passes through a characteristic extraction module consisting of three layers of convolutional neural networks, and the phase and frequency information of the signal contained in the sampling points are extracted and mapped into a characteristic vector with fixed dimension; the calculation is performed here using the one-dimensional convolution formula:
；
wherein,is the input signal sequence,/->Is a convolution filter, < >>Representing the result sequence of a one-dimensional convolution operation, +.>Is the index position in the output sequence, +.>Is a convolution kernel->Index position of->Element value representing the input signal,/->Weights representing the convolution kernel;
s3.2 building classification module
The feature vector is input to a classification module consisting of two stacked two-way long and short term memory neural networks, which can be expressed by the following formula:
；
；
；
；
；
；
in the above-mentioned formula(s),、/>、/>the outputs of the forget gate, the input gate and the output gate are respectively +.>Is an S-shaped growth curve, and the method comprises the following steps of,representing the result of a forgetting gate calculation +.>Is a weight matrix of forgetting gates, +.>Is the implicit state of the last time step, +.>Is the input of the current time step,/-, for example>Is a new candidate cell state, +.>And->Weight matrix of input gate and candidate cell state, respectively, +.>And->Is an input gate and a bias term for candidate cell states; />Representing time step->Is->Representing time step->In a module using a cell state, an implied state, a forget gate, an input gate and an output gate to control information flow;
s3.3 building Global perception Module
A global attention sensing module based on a multi-head attention mechanism is added between two stacked two-way long-short-period memory neural networks and is used for enhancing the capturing capacity of a model on global information; the calculation method of the module is as follows:
；
；
；
wherein,、/>and->Are linear transformations of the original input vector multiplied by a specific coefficient, < >>Represents->Sub-independent attention calculations, new output vector +.>，/>Connecting together a plurality of independent attention weighted sequence outputs, multiplying a weight to generate a final output +.>The method comprises the steps of carrying out a first treatment on the surface of the Each calculation will useFunction to generate an attention output +.>，/>The calculation formula of (2) is as follows:
；
wherein,representative vector->Is>Element(s)>Is->Probability of individual category->The exponential term representing all classes is summed for normalization, ensuring that one probability distribution is output, since the sum of all probability values is 1;
in the final part of the neural network model, use is made ofOutputting signal modulation mode category by the function, calculating multi-classification cross entropy loss to evaluate the prediction effect of the model, and optimizing model parameters by adopting a self-adaptive moment estimation algorithm;
s3.4 training a model using pre-generated underwater acoustic signals
Generating signal waveforms by using various modulation modes such as BPSK, QPSK, 8PSK, 2FSK, 4FSK, 2ASK, 16QAM and the like, taking the signal waveforms with the fixed dimension of 1 x 1000 as samples, and taking the corresponding modulation modes as labels to generate a data set; on the basis of the generated data set, adding a plurality of Gaussian white noises with different signal to noise ratios respectively to form a noisy data set; training the models by using various data sets without noise and with noise respectively, and selecting the model with highest accuracy as a final modulation recognition model.
6. The method for intelligently receiving underwater sound signals based on the noise reduction neural network according to claim 1, wherein the step S4 is specifically:
generating signal waveforms by utilizing multiple modulation modes respectively, taking the signal waveforms with fixed dimensions as samples, taking corresponding original symbols as labels, and generating a data set; on the basis of the generated data set, adding a plurality of Gaussian white noises with different signal to noise ratios respectively to form a noisy data set; training the models by utilizing a plurality of non-noisy and noisy data sets respectively, and selecting the model with the highest accuracy as a final demodulation model; and respectively training the demodulation models offline according to each modulation mode, and automatically selecting the demodulation models according to the output result of the modulation recognition model in a test stage to realize intelligent underwater sound signal receiving.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410063273.4A CN117614467B (en) | 2024-01-17 | 2024-01-17 | Underwater sound signal intelligent receiving method based on noise reduction neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410063273.4A CN117614467B (en) | 2024-01-17 | 2024-01-17 | Underwater sound signal intelligent receiving method based on noise reduction neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117614467A true CN117614467A (en) | 2024-02-27 |
| CN117614467B CN117614467B (en) | 2024-05-07 |
Family
ID=89954766
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410063273.4A Active CN117614467B (en) | 2024-01-17 | 2024-01-17 | Underwater sound signal intelligent receiving method based on noise reduction neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117614467B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| RU2681252C1 (en) * | 2018-04-16 | 2019-03-05 | Федеральное Государственное Казенное Военное Образовательное Учреждение Высшего Образования "Тихоокеанское Высшее Военно-Морское Училище Имени С.О. Макарова" Министерства Обороны Российской Федерации (Г. Владивосток) | Hydro acoustic signals detection and their neural network classification system |
| CN112887239A (en) * | 2021-02-15 | 2021-06-01 | 青岛科技大学 | Method for rapidly and accurately identifying underwater sound signal modulation mode based on deep hybrid neural network |
| CN113259288A (en) * | 2021-05-05 | 2021-08-13 | 青岛科技大学 | Underwater acoustic communication modulation mode identification method based on feature fusion and lightweight hybrid neural network |
| CN113269077A (en) * | 2021-05-19 | 2021-08-17 | 青岛科技大学 | Underwater acoustic communication signal modulation mode identification method based on improved gating network and residual error network |
| CN114897023A (en) * | 2022-05-19 | 2022-08-12 | 西安交通大学 | Underwater sound target identification method based on underwater sound target sensitivity difference feature extraction |
| US20220317273A1 (en) * | 2020-05-27 | 2022-10-06 | Soochow University | Underwater acoustic target ranging method based on feature extraction and neural network |
| US20230039900A1 (en) * | 2021-08-07 | 2023-02-09 | Fuzhou University | Method for realizing a multi-channel convolutional recurrent neural network eeg emotion recognition model using transfer learning |
| CN115982613A (en) * | 2022-08-31 | 2023-04-18 | 中国人民解放军空军工程大学 | Signal modulation identification system and method based on improved convolutional neural network |
| WO2023216583A1 (en) * | 2022-05-13 | 2023-11-16 | 河海大学 | Hail weather identification and classification method based on multi-channel deep residual shrinkage network |
| CN117081895A (en) * | 2023-10-16 | 2023-11-17 | 电子科技大学 | Automatic modulation identification method based on self-adaptive noise reduction |
| CN117115039A (en) * | 2023-09-08 | 2023-11-24 | 佛山读图科技有限公司 | Poisson diffusion model noise reduction method and system |
| CN117221053A (en) * | 2023-09-26 | 2023-12-12 | 中国人民解放军海军潜艇学院 | Underwater acoustic communication channel estimation method and system adopting improved DDPM noise reduction |
-
2024
- 2024-01-17 CN CN202410063273.4A patent/CN117614467B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| RU2681252C1 (en) * | 2018-04-16 | 2019-03-05 | Федеральное Государственное Казенное Военное Образовательное Учреждение Высшего Образования "Тихоокеанское Высшее Военно-Морское Училище Имени С.О. Макарова" Министерства Обороны Российской Федерации (Г. Владивосток) | Hydro acoustic signals detection and their neural network classification system |
| US20220317273A1 (en) * | 2020-05-27 | 2022-10-06 | Soochow University | Underwater acoustic target ranging method based on feature extraction and neural network |
| CN112887239A (en) * | 2021-02-15 | 2021-06-01 | 青岛科技大学 | Method for rapidly and accurately identifying underwater sound signal modulation mode based on deep hybrid neural network |
| CN113259288A (en) * | 2021-05-05 | 2021-08-13 | 青岛科技大学 | Underwater acoustic communication modulation mode identification method based on feature fusion and lightweight hybrid neural network |
| CN113269077A (en) * | 2021-05-19 | 2021-08-17 | 青岛科技大学 | Underwater acoustic communication signal modulation mode identification method based on improved gating network and residual error network |
| US20230039900A1 (en) * | 2021-08-07 | 2023-02-09 | Fuzhou University | Method for realizing a multi-channel convolutional recurrent neural network eeg emotion recognition model using transfer learning |
| WO2023216583A1 (en) * | 2022-05-13 | 2023-11-16 | 河海大学 | Hail weather identification and classification method based on multi-channel deep residual shrinkage network |
| CN114897023A (en) * | 2022-05-19 | 2022-08-12 | 西安交通大学 | Underwater sound target identification method based on underwater sound target sensitivity difference feature extraction |
| CN115982613A (en) * | 2022-08-31 | 2023-04-18 | 中国人民解放军空军工程大学 | Signal modulation identification system and method based on improved convolutional neural network |
| CN117115039A (en) * | 2023-09-08 | 2023-11-24 | 佛山读图科技有限公司 | Poisson diffusion model noise reduction method and system |
| CN117221053A (en) * | 2023-09-26 | 2023-12-12 | 中国人民解放军海军潜艇学院 | Underwater acoustic communication channel estimation method and system adopting improved DDPM noise reduction |
| CN117081895A (en) * | 2023-10-16 | 2023-11-17 | 电子科技大学 | Automatic modulation identification method based on self-adaptive noise reduction |
Non-Patent Citations (4)
| Title |
|---|
| ALEX NICHOL: ""Improved Denoising Diffusion Probabilistic Models"", ARXIV.ORG, 18 February 2021 (2021-02-18), pages 2 - 4 * |
| WEILONG ZHANG: "Modulation Recognition of Underwater Acoustic Signals Using Deep Hybrid Neural Networks", 《IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS》, 31 August 2022 (2022-08-31) * |
| 杨洁;夏卉;: "基于卷积神经网络的通信信号调制识别研究", 计算机测量与控制, no. 07, 25 July 2020 (2020-07-25) * |
| 邹友龙;胡法龙;李长喜;李潮流;: "利用径向基函数预测岩石渗透率及流体分子组分", 测井技术, no. 03, 20 June 2012 (2012-06-20) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117614467B (en) | 2024-05-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | A data preprocessing method for automatic modulation classification based on CNN | |
| CN113269077B (en) | Underwater acoustic communication signal modulation mode identification method based on improved gating network and residual error network | |
| CN112347871B (en) | Interference signal modulation identification method for communication carrier monitoring system | |
| CN110569752A (en) | convolutional neural network-based radar signal category determination method | |
| CN110690931B (en) | Digital signal adaptive code rate estimation method and device based on multi-wavelet-base combination | |
| CN112615804B (en) | Short burst underwater acoustic communication signal modulation identification method based on deep learning | |
| CN112910812B (en) | Modulation mode identification method for deep learning based on space-time feature extraction | |
| CN112737992B (en) | Underwater sound signal modulation mode self-adaptive in-class identification method | |
| CN113630130B (en) | End-to-end digital communication demodulation method | |
| Ali et al. | Automatic modulation classification using different neural network and PCA combinations | |
| CN116628566A (en) | Communication signal modulation classification method based on aggregated residual transformation network | |
| CN110233810A (en) | Msk signal demodulation method based on deep learning under a kind of mixed noise | |
| CN111612130B (en) | Frequency shift keying communication signal modulation mode identification method | |
| CN115982613A (en) | Signal modulation identification system and method based on improved convolutional neural network | |
| CN114615118A (en) | Modulation identification method based on multi-terminal convolution neural network | |
| Zhang et al. | Deep learning based underwater acoustic OFDM receiver with joint channel estimation and signal detection | |
| CN117614467B (en) | Underwater sound signal intelligent receiving method based on noise reduction neural network | |
| Zhang et al. | Efficient residual shrinkage CNN denoiser design for intelligent signal processing: Modulation recognition, detection, and decoding | |
| CN116094885A (en) | Signal modulation mode identification method based on One2ThreeNet | |
| Cao et al. | Learning to denoise and decode: A novel residual neural network decoder for polar codes | |
| CN115913849A (en) | Electromagnetic signal identification method based on one-dimensional complex value residual error network | |
| CN114584441A (en) | Digital signal modulation identification method based on deep learning | |
| Hussein et al. | Automatic Modulation Classification: Convolutional Deep Learning Neural Networks Approaches | |
| Zuo et al. | Convolutional Neural Network Modulation Recognition by Incorporating Attention Mechanisms | |
| CN114785649B (en) | Satellite communication signal identification method based on multiport neural network |
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 |