CN117131416A - Small sample modulation identification method, system, electronic equipment and storage medium - Google Patents

Small sample modulation identification method, system, electronic equipment and storage medium Download PDF

Info

Publication number
CN117131416A
CN117131416A CN202311052784.8A CN202311052784A CN117131416A CN 117131416 A CN117131416 A CN 117131416A CN 202311052784 A CN202311052784 A CN 202311052784A CN 117131416 A CN117131416 A CN 117131416A
Authority
CN
China
Prior art keywords
signal
layer
underwater communication
signals
model
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
Application number
CN202311052784.8A
Other languages
Chinese (zh)
Other versions
CN117131416B (en
Inventor
骆忠强
肖文诗
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sichuan University of Science and Engineering
Original Assignee
Sichuan University of Science and Engineering
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Sichuan University of Science and Engineering filed Critical Sichuan University of Science and Engineering
Priority to CN202311052784.8A priority Critical patent/CN117131416B/en
Publication of CN117131416A publication Critical patent/CN117131416A/en
Application granted granted Critical
Publication of CN117131416B publication Critical patent/CN117131416B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Molecular Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Computing Systems (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Digital Transmission Methods That Use Modulated Carrier Waves (AREA)

Abstract

The invention discloses a small sample modulation identification method, a system, electronic equipment and a storage medium, and relates to the technical field of modulation identification of communication signals. The method comprises the following steps: acquiring an underwater communication signal; carrying out signal reconstruction processing on the underwater communication signal to obtain an enhanced signal; determining a signal identification classification result according to the enhanced signal and the ResDLNN model; the ResDLNN model comprises an input layer, a residual error dense convolution block, a long-period memory layer and an output layer which are connected in sequence; the residual dense convolution block is constructed from two one-dimensional convolution layers with 128 kernels, one Dropout layer and one maximum pooling layer based on a residual network structure and a dense network structure; the residual dense convolution block is used for extracting features in the signal; the long-term and short-term memory layer is used for extracting time domain information in signal characteristics. The invention can improve the classification performance of the modulation signal on the basis of a small number of modulation signal samples by utilizing deep learning.

Description

Small sample modulation identification method, system, electronic equipment and storage medium
Technical Field
The present invention relates to the field of modulation recognition technology of communication signals, and in particular, to a method, a system, an electronic device, and a storage medium for recognizing modulation of a small sample.
Background
Automatic Modulation Recognition (AMR) is a promising wireless communication technology that can be implemented in internet of things and fifth generation wireless systems. Particularly in non-cooperative communication scenarios and systems with adaptive modulation techniques, the receiver first needs to identify the modulation type of the received signal, then can select an appropriate demodulation algorithm to complete demodulation, and finally obtains information. Thus, many scenarios require the use of automatic modulation recognition techniques to obtain information in the non-cooperative signal in order to complete the next work schedule.
In recent years, with the development of 5G technology, communication traffic has exploded, and conventional likelihood-based modulation recognition methods and feature-based modulation recognition methods have difficulty in accommodating such huge amounts of data. Therefore, the students and the experts in the communication field combine artificial intelligence with modulation recognition technology to solve the above-mentioned problems. At present, more and more scholars begin to work on researching a modulation recognition algorithm based on a deep learning technology, and the method not only can learn characteristics from huge communication signal data, but also can well finish classification tasks according to the characteristics, does not need a great deal of priori knowledge in the whole process, liberates human beings from the work of extracting the signal characteristics, and greatly improves the efficiency of the modulation recognition task.
Because the quantity of parameters in the neural network is huge, a small generalization error can be realized by large-scale training data, however, in certain fields, such as deep sea communication, satellite communication and radar communication scenes, a large amount of signal data is difficult to obtain, at this time, due to the fact that the quantity of data is not huge enough, the deep learning technology is difficult to learn relevant features from a small quantity of signal samples, further the generalization performance of the neural network is poor, and the accuracy of modulation recognition is greatly reduced. Therefore, it is a challenging task to efficiently utilize deep learning to improve the classification performance of modulated signals with less training data.
Disclosure of Invention
The invention aims to provide a small sample modulation identification method, a system, electronic equipment and a storage medium, which can improve the classification performance of a modulation signal on the basis of a small number of modulation signal samples by utilizing deep learning.
In order to achieve the above object, the present invention provides the following solutions:
a small sample modulation identification method, comprising:
acquiring an underwater communication signal;
performing signal reconstruction processing on the underwater communication signal to obtain an enhanced signal;
inputting the enhanced signal into a ResDLNN model for identification and classification to obtain a signal identification and classification result; the ResDLNN model comprises an input layer, a residual error dense convolution block, a long-period memory layer and an output layer which are connected in sequence; the residual dense convolution block is constructed by two one-dimensional convolution layers with 128 cores, one Dropout layer and one maximum pooling layer based on a residual network structure and a dense network structure; the input layer is used for receiving signals; the residual dense convolution block is used for extracting features in the signal; the long-term and short-term memory layer is used for extracting time domain information in signal characteristics; the output layer is used for classifying according to the time domain information.
Optionally, performing signal reconstruction processing on the underwater communication signal to obtain an enhanced signal, which specifically includes:
randomly shifting the underwater communication signal and adding noise to obtain a first reconstruction signal;
randomly cutting off the underwater communication signal and adding noise to obtain a second reconstruction signal;
and determining an enhancement signal according to the first reconstruction signal and the second reconstruction signal.
Optionally, performing random shift on the underwater communication signal and adding noise to obtain a first reconstruction signal, which specifically includes:
splitting the underwater communication signal into L sections, moving the last section of the L section signals to the forefront of the signals to form section shift signals, and finally randomly replacing one section of the section shift signals by Gaussian noise to obtain a first reconstruction signal.
Optionally, the underwater communication signal is randomly truncated and noise is added to obtain a second reconstructed signal, which specifically includes:
and cutting off the underwater communication signal, then splicing the cut-off part of the signal to obtain the whole signal, adding noise to realize signal reconstruction, and obtaining a second reconstruction signal.
Optionally, the training method of the ResDLNN model is as follows:
acquiring training data; the training data comprises signal data and corresponding signal classification results;
building a training model, setting the batch_size in the model to 400, the training times to 60 times, and the learning rate to 0.001;
and inputting the training data into the training model for training, and taking the trained model as a ResDLNN model.
The invention also provides a small sample modulation recognition system, which comprises:
the signal acquisition module is used for acquiring underwater communication signals;
the enhancement module is used for carrying out signal reconstruction processing on the underwater communication signals to obtain enhancement signals;
the signal identification module is used for inputting the enhanced signal into a ResDLNN model for identification and classification to obtain a signal identification and classification result; the ResDLNN model comprises an input layer, a residual error dense convolution block, a long-period memory layer and an output layer which are connected in sequence; the residual dense convolution block is constructed by two one-dimensional convolution layers with 128 cores, one Dropout layer and one maximum pooling layer based on a residual network structure and a dense network structure; the input layer is used for receiving signals; the residual dense convolution block is used for extracting features in the signal; the long-term and short-term memory layer is used for extracting time domain information in signal characteristics; the output layer is used for classifying according to the time domain information.
The invention also provides an electronic device comprising a memory for storing a computer program and a processor which runs the computer program to cause the electronic device to perform the small sample modulation recognition method according to the above.
The invention also provides a computer readable storage medium storing a computer program which when executed by a processor implements a small sample modulation recognition method as described above.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention discloses a small sample modulation identification method, a system, electronic equipment and a storage medium, wherein the method comprises the steps of obtaining underwater communication signals; carrying out signal reconstruction processing on the underwater communication signal to obtain an enhanced signal; determining a signal identification classification result according to the enhanced signal and the ResDLNN model; the ResDLNN model comprises an input layer, a residual error dense convolution block, a long-period memory layer and an output layer which are connected in sequence; the residual dense convolution block is constructed from two one-dimensional convolution layers with 128 kernels, one Dropout layer and one maximum pooling layer based on a residual network structure and a dense network structure; the residual dense convolution block is used for extracting features in the signal; the long-term and short-term memory layer is used for extracting time domain information in signal characteristics. The invention can realize data enhancement and transfer learning by utilizing deep learning, and improves the classification performance of the modulation signal on the basis of a small number of modulation signal samples.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a small sample modulation identification method of the present invention;
FIG. 2 is a schematic diagram of the underwater communication system in the present embodiment;
FIG. 3 is a random segmentation error diagram of the signal Duan Yiwei in the present embodiment;
FIG. 4 is a graph showing the variation of the 2FSK signal after the segment displacement and random segmentation error in the present embodiment;
FIG. 5 is a truncated random segmentation error map in this embodiment;
FIG. 6 is a graph showing the variation of the 4FSK signal after the segment displacement and random segmentation error in the present embodiment;
FIG. 7 is a diagram of a network configuration of ResDLNN in the present embodiment;
FIG. 8 is a diagram of a residual dense convolution block structure in this embodiment;
fig. 9 is a network configuration diagram of the LSTM in the present embodiment;
FIG. 10 is a diagram showing the comparison of the accuracy of modulation recognition of the ResDLNN model and other models in small sample underwater communication signal data in the present embodiment; wherein part (a) is a comparison plot of the underwater communication signal at 1% and part (b) is a comparison plot of the underwater communication signal at 3%;
FIG. 11 is a graph showing recognition accuracy and loss of each model before data enhancement and after data enhancement in 1% of the data of the entire underwater communication signal in the present embodiment; wherein, (a) is a contrast map of ResDLNN, (b) is a contrast map of CLDNN, (c) is a contrast map of Pro, (d) is a contrast map of SCRNN, and (e) is a contrast map of Transfer;
FIG. 12 is a graph showing the recognition accuracy and loss of each model before and after data enhancement in 3% of the data of the entire underwater communication signal in the present embodiment; wherein, (a) is a contrast chart of ResDLNN, (b) is a contrast chart of CLDNN, (c) is a contrast chart of Pro, (d) is a contrast chart of SCRNN, and (e) is a contrast chart of Transfer.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. 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 aims to provide a small sample modulation identification method, a system, electronic equipment and a storage medium, which can improve the classification performance of a modulation signal on the basis of a small number of modulation signal samples by utilizing deep learning.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
As shown in fig. 1, the present invention provides a small sample modulation identification method, which includes:
step 100: acquiring an underwater communication signal;
step 200: performing signal reconstruction processing on the underwater communication signal to obtain an enhanced signal;
step 300: inputting the enhanced signal into a ResDLNN model for identification and classification to obtain a signal identification and classification result; the ResDLNN model comprises an input layer, a residual error dense convolution block, a long-period memory layer and an output layer which are connected in sequence; the residual dense convolution block is constructed by two one-dimensional convolution layers with 128 cores, one Dropout layer and one maximum pooling layer based on a residual network structure and a dense network structure; the input layer is used for receiving signals; the residual dense convolution block is used for extracting features in the signal; the long-term and short-term memory layer is used for extracting time domain information in signal characteristics; the output layer is used for classifying according to the time domain information.
As a specific embodiment of step 200, specifically including:
step 201: randomly shifting the underwater communication signal and adding noise to obtain a first reconstruction signal; the specific process is as follows:
splitting the underwater communication signal into L sections, moving the last section of the L section signals to the forefront of the signals to form section shift signals, and finally randomly replacing one section of the section shift signals by Gaussian noise to obtain a first reconstruction signal.
Step 202: randomly cutting off the underwater communication signal and adding noise to obtain a second reconstruction signal; the specific process is as follows:
and cutting off the underwater communication signal, then splicing the cut-off part of the signal to obtain the whole signal, adding noise to realize signal reconstruction, and obtaining a second reconstruction signal.
Step 203: and determining an enhancement signal according to the first reconstruction signal and the second reconstruction signal.
The training method of the ResDLNN model comprises the following steps:
acquiring training data; the training data comprises signal data and corresponding signal classification results; building a training model, setting the batch_size in the model to 400, the training times to 60 times, and the learning rate to 0.001; and inputting the training data into the training model for training, and taking the trained model as a ResDLNN model.
On the basis of the technical scheme, the following embodiments are provided.
In this embodiment, in order to improve the classification performance of a modulation signal (for example, a deep sea communication modulation signal) in the case of a small number of modulation signal samples, a modulation recognition method based on data enhancement and migration learning is provided on the basis of the existing method. Because the modulation signal is actually a sampling result in a specific time period, the time period can be changed in actual operation, and noise is added to the signal when the receiving end samples the modulation signal, in order to simulate the actual situation, a certain part of the signal is randomly shifted, and meanwhile, the noise is added to reconstruct the signal, so that the data expansion and enhancement are completed. In addition, in the actual sampling process, sometimes the sampled signal is not complete, in order to simulate an actual scene, the original signal can be cut off, noise is added to realize signal reconstruction, and data enhancement of the modulated signal is realized. The transfer learning can find the similarity between the existing knowledge and the new knowledge, and the features of the new knowledge can be quickly learned through the similarity. Thus, transfer learning is often used in small sample learning tasks. Therefore, the combination of data enhancement and transfer learning can improve the generalization capability of deep learning under the condition of a small number of modulation signal samples, and further improve the recognition accuracy of the small sample modulation signals.
Firstly, carrying out random shift on an original underwater communication signal, and adding noise to complete signal reconstruction; in addition, the original underwater communication signals are required to be randomly truncated, and noise is added to complete signal reconstruction, so that data enhancement of a small number of underwater communication signals is completed. And then designing a new neural network model ResDLNN according to the existing deep neural network, applying the idea of transfer learning to automatic modulation recognition, pre-training the designed network model by taking a public data set RML2016.10A as a training sample, and finally learning features from the extended and enhanced underwater communication signals by utilizing the trained ResDLNN network to finally finish modulation recognition of the underwater communication signals.
The underwater communication signal is required to be sampled at a receiving end of an underwater communication system, then the underwater communication signal is subjected to data enhancement, so that a data set is expanded, a neural network-ResDLNN is better trained, then the ResDLNN is pre-trained by utilizing source domain data (RML20161.10A), and finally fine adjustment learning is performed on the pre-trained ResDLNN by utilizing the enhanced underwater communication signal, and finally the modulation recognition task of the underwater communication signal is completed.
1. Radio channel model and data set
1. Underwater signal communication system
The underwater signal communication system adopted by the invention is shown in figure 2, and comprises a transmitting end, a submarine communication channel and a receiving end. Firstly, converting a digital modulation signal into an analog signal by using a digital-to-analog converter, and amplifying the analog signal by using a power amplifier; the transmitting transducer then converts the analog signal into an acoustic signal for transmission; finally, the receiving end converts the sound signal into an analog signal by using the transducer, then filters the analog signal by using the filter, and finally samples the filtered analog signal by using the analog-to-digital converter, so as to obtain an underwater communication signal data set.
The underwater communication signal received by the receiving end can be expressed by the following formula (1):
r(t)＝f(s(t))+n(t) (1)
wherein r (t) refers to a signal received by a receiving end, s (t) refers to a modulation signal sent by a sending end, f (·) represents an underwater sound channel, and n (t) represents noise in the channel.
2. Data set
In 2020, the Qingdao university related subject group collects underwater communication signal data sets, namely yellow sea data sets, in the shallow offshore area of the yellow sea of Qingdao city, shandong province, the water depth of the collected underwater sound signal data sets is 4.2 meters, the sampling rate of the digital-to-analog converter is 336KHz, the sampling rate of the analog-to-digital converter is 336KHz, and the modulation modes of the transmitted signals are 2FSK, 4FSK, 2PSK, 4PSK, 16QAM,64QAM and OFDM. Each type of modulation mode has 200 signals, the sampling point of each signal is 50000, and 1400 signals are all used. To make an in-water communication signal with a small number of samples, a signal with a signal length of 50000 is truncated into a signal with a signal length of 128. Simultaneously, 1400 signals with the length of 50000 are adjusted to 5600 signals with the length of 128, and the 5600 signals are taken as 1% of the whole water communication signal data; in addition, 1400 signals with the length of 50000 are adjusted to 15400 signals with the length of 128, and the 15400 signals are taken as 3% of the whole underwater communication signal data. And then converting the underwater signal into a complex signal by using Hilbert transformation, separating the real part and the imaginary part of the complex signal, and forming the real part and the imaginary part into signal samples in an IQ data format, wherein the dimension of each signal is.
In 2016, O' Shea, T.J introduced in Convolutional Radio Modulation Recognition Networks a modulation identification data set rml2016.10a, which is an analog and digital signal of 11 modulation types (8 psk, bpsk, cpfsk, gfsk, pam4,16qam,64qam, qpsk, AM-DSB, AM-SSB, WBFM) acquired using GNU Radio software connectivity software Radio peripherals. The channel simulation takes into account the effects of additive white gaussian noise, selective fading, center frequency offset, sample rate offset, etc. The signal-to-noise ratio range is-20 dB-18 dB, samples of the data set contain IQ two paths of components, the sampling points of each path are 128, and the data set contains 22 ten thousands of sample signals with IQ data formats.
2. Underwater acoustic signal data enhancement
1. Random segment error of segment shift
In an actual receiving end, the modulation signal is actually a sampling result in a certain period of time, and the sampling period of time can be arbitrarily changed; meanwhile, an uncertainty error may be generated in the actual sampling process. In order to simulate the actual situation, a technique of shifting a signal segment by a random segmentation error is proposed, wherein the technique is to randomly divide a modulated signal into L segments, shift the last segment of the signal to the forefront of the signal to form a segment shift signal, and finally randomly replace one segment of the segment shift signal by Gaussian noise. The signal segment shift + random segment error operation can be written as:
wherein, r (t)' represents a new signal after segment shift random segment error operation, r i Representing a certain segment in the original signal, L representing the length of the original signal, noise representing gaussian noise. The visual representation of this process is shown in fig. 3.
As shown in fig. 3, the signal is first divided into L segments, the length of each segment is randomly selected, then the last segment is shifted to the first segment before the last segment is shifted to the first segment, then the last segment is shifted to the first segment on the basis, and finally a segment of signal is randomly replaced by gaussian noise. The comparison of the 2fsk signal before and after the segment shift random segmentation error operation is shown in fig. 4.
2. Truncated random segmentation error
In the actual sampling process, sometimes the sampled signal is not complete, in order to splice out the complete modulated signal, the original signal may be subjected to a truncation operation, then the truncated portion of the signal is spliced out of the whole signal, and noise is added to realize signal reconstruction. The signal truncation + random segment error operation may be written as:
r(t)″＝[r a ,r a +n,...,noise,r b +m,…,r c +o] (3)
where r (t) "represents the new signal after the truncated random segmentation error operation, a, b, c represents the start position of the phase, n, m, o represents the length of the truncation, and these values are random. The visual representation of this process is shown in fig. 5.
As shown in fig. 5, the signal is truncated first, the starting position and the truncation length of the truncation are random, then the truncated signals are spliced together, and gaussian noise is used to replace a certain segment of the signals, so that the comparison of the signal reconstruction 4fsk signal before and after the truncation random segmentation error operation is completed is shown in fig. 6.
3. Designing a network model
The ResDLNN model structure designed in this embodiment is shown in fig. 7, and the corresponding parameters are listed in table 1. The network model can be divided into four parts: an input layer, a residual dense convolution block, a long short term memory Layer (LSTM), and an output layer. The details of each part are as follows:
TABLE 1 ResDLNN network architecture parameters
As shown in table 1, the proposed network of the present invention includes Input layers (Input), 4 convolutional layers, 2 LSTM layers and one full connection layer. The Input layer is used for receiving signals, and the Input layer converts signal data with the dimension of 2×128 into 3 dimensions, namely (batch_size, input_size), wherein the input_size is the dimension of the modulated signal of 2×128, and the batch_size refers to the number of 2×128 (i.e. how many signals with the dimension of 2×128). The convolution layer is used for extracting features in signals, the Dropout layer is used for solving the overfitting phenomenon of the neural network, and the Add layer can perform 'adding' operation on the output features of the previous layers and the output features of the current layer so as to reduce the loss of the features. The LSTM layer is used for extracting time domain features in the modulated signals, so that the features extracted by the whole network are richer, and the classification of the full connection layer is facilitated. The Dense layer can calculate the probability of each modulation signal according to the characteristics extracted by the network, and finally output the modulation type with the maximum probability, so as to judge the modulation mode of the current input signal.
(1) Input layer: because the invention is used for transfer learning, the designed neural network is trained by using the signals in the source domain, and then the signals in the target domain are input into the pre-trained neural network for fine tuning. Thus, the signals in the source domain and the signals in the target domain commonly use the input layer of the network model, so that the sizes of the input signals in the source domain and the target domain need to be matched. The dimension of the signal in the source domain is 2×128, and the dimension of the underwater sound signal in the target domain is 1×50000, so the invention cuts off the underwater sound signal to form a signal with the dimension of 1×128. The signal in the source domain is a 2 x 128 matrix formed by combining the real and imaginary parts of the signal, so the invention first performs a hilbert transform on the underwater acoustic signal in the target domain, converts it into a complex signal, then separates the real and imaginary parts of the signal, and combines the real and imaginary parts to form a matrix of dimension 2 x 128.
(2) Residual dense convolution block
The proposed residual dense convolution block combines a residual network structure and a dense network structure, and the network structure of the residual dense convolution block is shown in fig. 8. The residual dense convolution block consists of 2 one-dimensional convolution layers with 128 cores, a Dropout layer and a maximum pooling layer, and the module introduces a dense connection mode in a dense connection network structure to enhance the reuse and gradient flow of the features, and meanwhile, in order to avoid the problem of feature reduction caused in the information forward transmission process, the invention is influenced by the residual network structure, and performs unit addition operation on the feature information of the previous layers and the feature information extracted by the later convolution layers, so that the network can learn better features.
(3) Long and short term memory layer
As described above, the residual dense convolution block can improve the propagation of the modulated signal features in the deep neural network, effectively solve the problems of overfitting and feature reduction, and can extract the features of the signal from space. The long and short term memory network is called a cell with memory cells, is good at processing time series data, modulates signals to be sampled in a certain time period, and is therefore time series data.
The network structure of LSTM is shown in FIG. 9, f t I is a forgetful door t For input door, C t In the state of a cell, the cell is in a state of being,is a candidate value of cell state, h t Is a hidden layer state value. The expressions of the forgetting gate, the input gate, the cell state candidate value, the hidden state value and the output gate at the time t in the LSTM are shown in formulas (4) - (9):
f t ＝σ(W f ·[h t-1 ,x t ]+b f ) (4)
i t ＝σ(W i ·[h t-1 ,x t ]+b i ) (5)
h t ＝o t tanh(c t ) (8)
h t ＝σ(W o ·[h t-1 ,x t ]+b o ) (9)
where σ is a sigmoid function, W is a weight, and b is a bias.
According to the above description, the input gate and the tanh of the LSTM may determine which information is acquired from the previous time, and may also combine the forgetting gate and the input gate to implement information discarding and saving and obtain the cell state at the current time, and finally obtain the hidden layer state at the current time. Since LSTM has forgetting function, it can reduce the possibility of gradient disappearance and gradient explosion, and solve the problem of long-term dependence which RNN cannot solve.
(4) Output layer
The second layer LSTM converts the multi-dimensional input to a one-dimensional output and then inputs the one-dimensional output into the fully connected layer. The parameters of the full connection layer are set to M (M represents the kind of the modulation signal), and the SoftMax function is used as an activation function, thereby outputting the probability of each modulation type, and finally determining the modulation type of the signal by outputting the probability value of the layer.
4. Small sample modulation recognition method based on data enhancement and transfer learning
In order to improve the classification performance of a modulation signal (such as a deep sea communication modulation signal) under the condition of a small number of modulation signal samples, the invention applies the idea of transfer learning to automatic modulation recognition, pre-trains a deep neural network through a plurality of training samples, then carries out fine adjustment by using the small sample modulation signal, and finally completes the modulation recognition under the condition of the small sample, and the specific steps are as follows:
(1) Firstly, a migration learning model is provided for an automatic modulation recognition task, and parameters are set. The proposed neural network-ResDLNN is then trained using the signals in the public dataset RML2016.10A as the source domain.
(2) The underwater communication signals of the target domain are subjected to data expansion by using a segment shift random segmentation error and truncated random segmentation error technology, and 5600 underwater communication signals with 2×128 dimensions are expanded to 16800 signals with 2×128 dimensions.
(3) 60% of the extended underwater communication signals are used as training sets, 20% are used as verifiers, and 20% are used as test sets. The training set is then input into the pre-trained ResDLNN for retraining. In the training process, the batch_size is set to 400, the training times is set to 60, the learning rate is set to 0.001, and the optimization algorithm is set to Adam.
In order to better illustrate the technical effects of the invention, the following experiments were performed: (1) In order to verify the classification performance of the ResDLNN model provided by the invention under the condition of further reducing training samples, under the condition that the training samples are 1% of all underwater communication signals and the training samples are 3% of all underwater communication signals, a comparison experiment is conducted on the classification accuracy of the ResDLNN model and the existing common network model for AMR. FIG. 10 is a graph of modulation recognition accuracy versus various neural networks. (2) In order to verify the data enhancement method provided by the invention, the modulation recognition accuracy of each model before data enhancement is compared with that of each model after data enhancement, and fig. 11 and fig. 12 are comparison diagrams of the modulation recognition accuracy of each network model before and after data enhancement. Through the two experiments, the ResDLNN model provided by the invention has more excellent recognition accuracy than other modulation recognition models, and meanwhile, the data enhancement method provided by the invention can improve the modulation recognition accuracy. In summary, the modulation recognition method based on data enhancement and transfer learning provided by the invention can improve the recognition rate of the signal modulation mode under the condition of a small number of modulation signal samples.
In addition, the invention also provides a small sample modulation recognition system, which comprises:
the signal acquisition module is used for acquiring underwater communication signals;
the enhancement module is used for carrying out signal reconstruction processing on the underwater communication signals to obtain enhancement signals;
the signal identification module is used for inputting the enhanced signal into a ResDLNN model for identification and classification to obtain a signal identification and classification result; the ResDLNN model comprises an input layer, a residual error dense convolution block, a long-period memory layer and an output layer which are connected in sequence; the residual dense convolution block is constructed by two one-dimensional convolution layers with 128 cores, one Dropout layer and one maximum pooling layer based on a residual network structure and a dense network structure; the input layer is used for receiving signals; the residual dense convolution block is used for extracting features in the signal; the long-term and short-term memory layer is used for extracting time domain information in signal characteristics; the output layer is used for classifying according to the time domain information.
The invention also provides an electronic device comprising a memory for storing a computer program and a processor which runs the computer program to cause the electronic device to perform the small sample modulation recognition method according to the above.
The invention also provides a computer readable storage medium storing a computer program which when executed by a processor implements a small sample modulation recognition method as described above.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other.
The principles and embodiments of the present invention have been described herein with reference to specific examples, the description of which is intended only to assist in understanding the core concept of the invention; also, it is within the scope of the present invention to be modified by those of ordinary skill in the art in light of the present teachings. In view of the foregoing, this description should not be construed as limiting the invention.

Claims (8)

1. A small sample modulation identification method, comprising:
acquiring an underwater communication signal;
performing signal reconstruction processing on the underwater communication signal to obtain an enhanced signal;
inputting the enhanced signal into a ResDLNN model for identification and classification to obtain a signal identification and classification result; the ResDLNN model comprises an input layer, a residual error dense convolution block, a long-period memory layer and an output layer which are connected in sequence; the residual dense convolution block is constructed by two one-dimensional convolution layers with 128 cores, one Dropout layer and one maximum pooling layer based on a residual network structure and a dense network structure; the input layer is used for receiving signals; the residual dense convolution block is used for extracting features in the signal; the long-term and short-term memory layer is used for extracting time domain information in signal characteristics; the output layer is used for classifying according to the time domain information.
2. The small sample modulation and identification method according to claim 1, wherein the signal reconstruction processing is performed on the underwater communication signal to obtain an enhanced signal, specifically comprising:
randomly shifting the underwater communication signal and adding noise to obtain a first reconstruction signal;
randomly cutting off the underwater communication signal and adding noise to obtain a second reconstruction signal;
and determining an enhancement signal according to the first reconstruction signal and the second reconstruction signal.
3. The method for identifying small sample modulation according to claim 2, wherein the step of randomly shifting the underwater communication signal and adding noise to obtain the first reconstructed signal comprises:
splitting the underwater communication signal into L sections, moving the last section of the L section signals to the forefront of the signals to form section shift signals, and finally randomly replacing one section of the section shift signals by Gaussian noise to obtain a first reconstruction signal.
4. The small sample modulation and identification method according to claim 2, wherein the step of randomly truncating the underwater communication signal and adding noise to obtain a second reconstructed signal comprises:
and cutting off the underwater communication signal, then splicing the cut-off part of the signal to obtain the whole signal, adding noise to realize signal reconstruction, and obtaining a second reconstruction signal.
5. The small sample modulation recognition method of claim 1, wherein the training method of the ResDLNN model is as follows:
acquiring training data; the training data comprises signal data and corresponding signal classification results;
building a training model, setting the batch_size in the model to 400, the training times to 60 times, and the learning rate to 0.001;
and inputting the training data into the training model for training, and taking the trained model as a ResDLNN model.
6. A small sample modulation recognition system, comprising:
the signal acquisition module is used for acquiring underwater communication signals;
the enhancement module is used for carrying out signal reconstruction processing on the underwater communication signals to obtain enhancement signals;
the signal identification module is used for inputting the enhanced signal into a ResDLNN model for identification and classification to obtain a signal identification and classification result; the ResDLNN model comprises an input layer, a residual error dense convolution block, a long-period memory layer and an output layer which are connected in sequence; the residual dense convolution block is constructed by two one-dimensional convolution layers with 128 cores, one Dropout layer and one maximum pooling layer based on a residual network structure and a dense network structure; the input layer is used for receiving signals; the residual dense convolution block is used for extracting features in the signal; the long-term and short-term memory layer is used for extracting time domain information in signal characteristics; the output layer is used for classifying according to the time domain information.
7. An electronic device comprising a memory for storing a computer program and a processor that runs the computer program to cause the electronic device to perform the small sample modulation recognition method according to claims 1-5.
8. A computer-readable storage medium, characterized in that it stores a computer program which, when executed by a processor, implements the small sample modulation recognition method as claimed in claims 1-5.
CN202311052784.8A 2023-08-21 2023-08-21 Small sample modulation identification method, system, electronic equipment and storage medium Active CN117131416B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311052784.8A CN117131416B (en) 2023-08-21 2023-08-21 Small sample modulation identification method, system, electronic equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311052784.8A CN117131416B (en) 2023-08-21 2023-08-21 Small sample modulation identification method, system, electronic equipment and storage medium

Publications (2)

Publication Number Publication Date
CN117131416A true CN117131416A (en) 2023-11-28
CN117131416B CN117131416B (en) 2024-06-04

Family

ID=88859298

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311052784.8A Active CN117131416B (en) 2023-08-21 2023-08-21 Small sample modulation identification method, system, electronic equipment and storage medium

Country Status (1)

Country Link
CN (1) CN117131416B (en)

Citations (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112054979A (en) * 2020-09-14 2020-12-08 四川大学 Radio automatic modulation identification method based on fuzzy dense convolution network
CN112241724A (en) * 2020-10-30 2021-01-19 南京信息工程大学滨江学院 Automatic identification method and system based on double-path convolution long-term and short-term neural network
CN112329523A (en) * 2020-09-25 2021-02-05 泰山学院 Underwater acoustic signal type identification method, system and equipment
CN112418014A (en) * 2020-11-09 2021-02-26 南京信息工程大学滨江学院 Modulation signal identification method based on wavelet transformation and convolution long-short term memory neural network
CN112702294A (en) * 2021-03-24 2021-04-23 四川大学 Modulation recognition method for multi-level feature extraction based on deep learning
CN112836569A (en) * 2020-12-15 2021-05-25 泰山学院 Underwater acoustic communication signal identification method, system and equipment based on sequence convolution network
CN113947151A (en) * 2021-10-20 2022-01-18 嘉兴学院 Automatic modulation and identification method for wireless communication signals in offshore complex environment
CN114239749A (en) * 2021-12-23 2022-03-25 桂林电子科技大学 Modulation identification method based on residual shrinkage and bidirectional long-short term memory network
CN114422311A (en) * 2022-01-26 2022-04-29 中国人民解放军战略支援部队信息工程大学 Signal modulation identification method and system combining deep neural network and expert prior characteristics
CN114595729A (en) * 2022-04-01 2022-06-07 吉林大学 Communication signal modulation identification method based on residual error neural network and meta-learning fusion
CN114764577A (en) * 2022-04-29 2022-07-19 天津大学 Lightweight modulation recognition model based on deep neural network and method thereof
CN114881092A (en) * 2022-06-16 2022-08-09 杭州电子科技大学 Signal modulation identification method based on feature fusion
CN114912489A (en) * 2022-05-09 2022-08-16 中国人民解放军空军工程大学 Signal modulation identification method
CN115834310A (en) * 2023-02-15 2023-03-21 四川轻化工大学 Communication signal modulation identification method based on LGTransformer
CN116319210A (en) * 2023-03-23 2023-06-23 中国人民解放军战略支援部队信息工程大学 Signal lightweight automatic modulation recognition method and system based on deep learning

Patent Citations (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112054979A (en) * 2020-09-14 2020-12-08 四川大学 Radio automatic modulation identification method based on fuzzy dense convolution network
CN112329523A (en) * 2020-09-25 2021-02-05 泰山学院 Underwater acoustic signal type identification method, system and equipment
CN112241724A (en) * 2020-10-30 2021-01-19 南京信息工程大学滨江学院 Automatic identification method and system based on double-path convolution long-term and short-term neural network
CN112418014A (en) * 2020-11-09 2021-02-26 南京信息工程大学滨江学院 Modulation signal identification method based on wavelet transformation and convolution long-short term memory neural network
CN112836569A (en) * 2020-12-15 2021-05-25 泰山学院 Underwater acoustic communication signal identification method, system and equipment based on sequence convolution network
CN112702294A (en) * 2021-03-24 2021-04-23 四川大学 Modulation recognition method for multi-level feature extraction based on deep learning
CN113947151A (en) * 2021-10-20 2022-01-18 嘉兴学院 Automatic modulation and identification method for wireless communication signals in offshore complex environment
CN114239749A (en) * 2021-12-23 2022-03-25 桂林电子科技大学 Modulation identification method based on residual shrinkage and bidirectional long-short term memory network
CN114422311A (en) * 2022-01-26 2022-04-29 中国人民解放军战略支援部队信息工程大学 Signal modulation identification method and system combining deep neural network and expert prior characteristics
CN114595729A (en) * 2022-04-01 2022-06-07 吉林大学 Communication signal modulation identification method based on residual error neural network and meta-learning fusion
CN114764577A (en) * 2022-04-29 2022-07-19 天津大学 Lightweight modulation recognition model based on deep neural network and method thereof
CN114912489A (en) * 2022-05-09 2022-08-16 中国人民解放军空军工程大学 Signal modulation identification method
CN114881092A (en) * 2022-06-16 2022-08-09 杭州电子科技大学 Signal modulation identification method based on feature fusion
CN115834310A (en) * 2023-02-15 2023-03-21 四川轻化工大学 Communication signal modulation identification method based on LGTransformer
CN116319210A (en) * 2023-03-23 2023-06-23 中国人民解放军战略支援部队信息工程大学 Signal lightweight automatic modulation recognition method and system based on deep learning

Non-Patent Citations (10)

* Cited by examiner, † Cited by third party
Title
MINGCHUN LI等: "Blind Separation for Wireless Communication Convolutive Mixtures Based on Denoising IVA", 《IEEE ACCESS》, 1 November 2022 (2022-11-01), pages 113756 - 113766 *
WENSHI XIAO等: "A Review of Research on Signal Modulation Recognition Based on Deep Learning", 《MDPI》, 1 September 2022 (2022-09-01), pages 1 - 29 *
侯涛等: "基于深度学习的通信信号调制方式识别", 《无线电工程》, 31 December 2019 (2019-12-31), pages 796 - 800 *
吴傲: "基于大数据的信号调制方式识别研究及软件设计", 《中国优秀硕士学位论文全文数据库信息科技辑》, 15 September 2019 (2019-09-15), pages 136 - 444 *
李纵宇: "基于深度学习的通信信号调制识别", 《中国优秀硕士学位论文全文数据库 信息科技辑》, 15 April 2022 (2022-04-15), pages 136 - 154 *
杨洪娟等: "基于联合特征参数的卫星单-混信号调制识别研究", 《电子与信息学报》, 31 October 2022 (2022-10-31), pages 3499 - 3506 *
王春: "基于卷积神经网络的数字信号调制识别研究", 《中国优秀硕士学位论文全文数据库信息科技辑》, 15 February 2020 (2020-02-15), pages 136 - 240 *
谷万博: "针对调制识别的高准确率低复杂度的深度学习算法", 《中国优秀硕士学位论文全文数据库信息科技辑》, 15 August 2023 (2023-08-15), pages 136 - 33 *
陈志运: "基于USRP和深度学习的电磁干扰信号检测与识别***", 《中国优秀硕士学位论文全文数据库信息科技辑》, 15 January 2022 (2022-01-15), pages 135 - 27 *
黄子豪: "水声信号调制方式智能识别方法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》, 15 January 2023 (2023-01-15), pages 136 - 1324 *

Also Published As

Publication number Publication date
CN117131416B (en) 2024-06-04

Similar Documents

Publication Publication Date Title
Liu et al. Deep neural network architectures for modulation classification
CN112418014B (en) Modulated signal identification method based on wavelet transformation and convolution long-term and short-term memory neural network
CN112464837B (en) Shallow sea underwater acoustic communication signal modulation identification method and system based on small data samples
CN110598677B (en) Space-time multi-channel deep learning system for automatic modulation recognition
CN110532932B (en) Method for identifying multi-component radar signal intra-pulse modulation mode
CN112615804A (en) Short burst underwater acoustic communication signal modulation identification method based on deep learning
CN112466320A (en) Underwater acoustic signal noise reduction method based on generation countermeasure network
CN113298031B (en) Signal modulation identification method and system considering signal physical and time sequence characteristics
CN111431825B (en) Signal automatic classification and identification method based on deep multi-flow neural network
CN114492522B (en) Automatic modulation classification method based on improved stacked hourglass neural network
CN112836569A (en) Underwater acoustic communication signal identification method, system and equipment based on sequence convolution network
CN114595729A (en) Communication signal modulation identification method based on residual error neural network and meta-learning fusion
CN114548201B (en) Automatic modulation identification method and device for wireless signal, storage medium and equipment
CN113902095A (en) Automatic modulation identification method, device and system for wireless communication
CN117131416B (en) Small sample modulation identification method, system, electronic equipment and storage medium
CN113743317A (en) License plate picture generation method, medium and equipment based on generation countermeasure network
CN116132235B (en) Continuous phase modulation signal demodulation method based on deep learning
CN112348165A (en) Underwater acoustic communication signal classification and identification method and system based on hybrid cycle network
Kalade et al. Using sequence to sequence learning for digital bpsk and qpsk demodulation
CN113343924B (en) Modulation signal identification method based on cyclic spectrum characteristics and generation countermeasure network
CN112580598B (en) Radio signal classification method based on multichannel Diffpool
CN115955375A (en) Modulated signal identification method and system based on CNN-GRU and CA-VGG feature fusion
Dong et al. Innovative semantic communication system
CN116415653A (en) Class increment neural network model aggregation method based on knowledge distillation
Wang et al. Modulation recognition method for underwater acoustic communication signal based on relation network under small sample set

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant