Usrp modulation signal recognition method based on migration deep learning

Abstract

The application relates to the technical field of software radio transmission, and discloses a USRP modulation signal identification method based on migration deep learning, wherein wireless signals are transmitted through a transmitting end, a USRP receiving end stores signal reception through a signal acquisition program after receiving the signals, and provides a required signal training set and a signal test set for a CNN model for modulation signal identification, wherein the signal training set comprises a pre-training data set and a migration learning training data set; the pre-training data set is generated, and a convolutional neural network model for modulation signal identification is pre-trained; the migration learning training data set is generated, and the convolutional neural network model for modulation signal identification is pre-trained; the trained migration learning convolutional neural network model is deployed to the signal test set, and a modulation signal identification result is verified; and the application solves the problem of modulation signal identification in wireless communication.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of software radio transmission, and in particular to a USRP modulation signal recognition method based on migration deep learning. BACKGROUND

[0002] Modulation recognition technology has important application value in both commercial and military fields. Modulation recognition technology can be divided into modulation recognition based on likelihood ratio and modulation recognition based on signal features. Modulation recognition based on likelihood ratio uses the principle of statistical inference, compares the likelihood ratio of known modulation types by constructing a likelihood ratio model of modulation types, compares the likelihood ratio of various modulation types with the signal to be identified, and determines the modulation type of the unknown signal. Modulation recognition based on signal features is to extract feature parameters or feature sequences of the signal (including spectral characteristics, power spectral density, phase change, instantaneous amplitude, modulation index, etc.), and then determine the modulation type of the signal according to these features.

[0003] Because the method based on likelihood ratio needs to accurately establish the likelihood ratio model of the modulation type, it may face challenges in model construction for complex modulation types. The method based on signal features is more flexible and can adapt to different modulation types according to different feature parameters, but it needs appropriate feature extraction and decision rule design, so the modulation recognition technology based on signal features is usually more widely applied. The traditional modulation recognition technology based on signal features first designs and extracts signal features, and then designs classification rules to classify the modulation type of the signal features. The manually designed signal features and classification rules usually have limited recognition accuracy in complex channel environments. In recent years, the development of deep learning has promoted its wide application in the fields of computer vision, natural language processing, etc., and has achieved remarkable results. Therefore, deep learning is introduced into modulation recognition technology. Deep learning automatically learns classification rules from feature data, thereby improving classification accuracy. Deep learning algorithms can automatically learn signal features and classification rules in the field of modulation recognition, thereby further improving the performance of modulation recognition.

[0004] Compared with traditional schemes, the modulation signal scheme based on deep learning has its advantages, especially in signal feature processing. At the same time, this scheme also has some shortcomings, and the convolutional neural network needs to be effectively trained, which puts certain requirements on the scale of the training set. For small-scale data training scenarios, the recognition performance is limited. In addition, when the channel environment changes, the trained model cannot meet the performance requirements of modulation signal recognition. In this case, a modulation signal recognition scheme based on deep migration learning is proposed to solve these scene defects. SUMMARY

[0005] The application aims to solve the above problems, and designs a USRP modulation signal recognition method based on migration deep learning.

[0006] The application provides a USRP modulation signal recognition method based on migration deep learning, which comprises the following steps:

[0007] Step 1, signal data acquisition: wireless signals are transmitted by a transmitting end, and after receiving the signals, a USRP receiving end stores the signal reception through a signal acquisition program, and provides a required signal training set and a signal test set for a CNN model for modulation signal recognition, wherein the signal training set comprises a pre-training data set and a migration learning training data set;

[0008] Step 2, pre-training of a convolutional neural network model: the pre-training data set is generated, and a convolutional neural network model for modulation signal recognition is pre-trained;

[0009] Step 3, migration learning training of the convolutional neural network model: the migration learning training data set is generated, and the convolutional neural network model for modulation signal recognition is pre-trained;

[0010] Step 4: the trained migration learning convolutional neural network model is deployed to the signal test set, and the modulation signal recognition result is verified.

[0011] Optionally, in the first implementation manner of the application, the radio frequency parameters of the transmitting end in step 1 at least comprise a radio frequency antenna, a center frequency, a transmitting gain and a modulation mode, and the radio frequency parameters of the USRP receiving end at least comprise a radio frequency antenna, a center frequency and a receiving gain.

[0012] Optionally, in the second implementation manner of the application, in step 1, the USRP receiving end starts receiving wireless information from a receiving antenna, and after passing through an RF receiving module, the modulation signal is received after analog-digital conversion and down-conversion processing.

[0013] Optionally, in the third implementation manner of the application, in step 1, the pre-training data set: the classified signal set adopted in a channel environment I is Ω4={BPSK, 8PSK, 4QAM, 16QAM}, the signal sequence length is L=100, the average signal-to-noise ratio of the sequence is uniformly distributed between [-10dB, 20dB], the number of each signal is N I =100000, the pre-training data set is set as N signals of each type, a total of 4N signal sequences, 90%×4N is taken as a training set in training, and 10%×4N is used for cross-validation, the test set is 1000 for each signal type, the SNR range is [-10dB, 20dB], testing is performed every interval, and the average is taken to obtain the modulation signal recognition rate.

[0014] Optionally, in the fourth implementation mode of the present application, the step 1 of migrating learning training dataset: in the channel environment II, the number of each signal type in the migrating learning training set is set to N II In the test set, there are 1000 signal types for each signal type, the SNR range is [-10dB, 20dB], the test is performed every 2dB interval, and the average result is taken to obtain the modulation signal recognition rate.

[0015] Optionally, in the fifth implementation mode of the present application, the step 3 specifically comprises: the USRP receiver collects signal data sets in two different channel environments, and adopts the migrating learning convolutional neural network to perform modulation signal recognition, and the migrating training is performed on the convolutional layer and the fully connected layer on the basis of the convolutional neural network.

[0016] Optionally, in the sixth implementation mode of the present application, the migrating learning convolutional neural network adopts two convolutional layers, the convolutional step length is 1, the first layer is a 64-core convolutional layer, the size of the first convolutional kernel is 2x4, the excitation function is ReLU, the bias term is adopted and zero value initialization is performed, the weight initial method is Glorot uniform distribution initialization, the number of the second convolutional kernel is 16, the size of the convolutional kernel is 1x4, and the remaining settings are the same as those of the first convolutional layer.

[0017] Optionally, in the seventh implementation mode of the present application, the migrating learning convolutional neural network fully connected neural network comprises three fully connected layers, the first two layers are respectively set to 64 and 16 neurons, the activation function is ReLU, the weight is initialized by Glorot uniform distribution, and the regularization term and weight constraint are adopted. The output layer activation function adopts Softmax.

[0018] In the technical scheme provided by the application, the transmission is performed by the transmitting end of a wireless signal, and after the signal is received by the USRP receiving end, the signal is received and stored by a signal acquisition program, so as to provide the required signal training set and signal test set for the CNN model for modulated signal identification, in the process, the signal training set includes a pre-training data set and a migration learning training data set, first, the pre-training data set is generated, and the convolutional neural network model for modulated signal identification is pre-trained to improve the initial performance of the model, then, the migration learning training data set is generated, and the convolutional neural network model for modulated signal identification is further pre-trained to adapt to specific modulated signal identification tasks, finally, the trained migration learning convolutional neural network model is deployed on the signal test set, and the modulated signal identification result is verified to ensure the accuracy and reliability of the model, the communication test environment based on the USRP software radio is built, the signal modulation identification technology based on deep learning is successfully verified to effectively solve the modulated signal identification problem in wireless communication, and the great potential and advantages of the signal modulation identification technology based on deep learning in practical application are displayed. BRIEF DESCRIPTION OF DRAWINGS

[0019] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are included only to illustrate preferred embodiments and do not imply any limitation on the application.

[0020] Figure 1 A USRP modulated signal identification method based on migration deep learning provided for an embodiment of the application is shown in the schematic diagram.

[0021] Figure 2 A transmitting end functional block diagram provided for an embodiment of the application is shown in the schematic diagram.

[0022] Figure 3 A receiving end functional block diagram provided for an embodiment of the application is shown in the schematic diagram.

[0023] Figure 4 A CNN deep learning neural network structure schematic diagram provided for an embodiment of the application is shown in the schematic diagram.

[0024] Figure 5 A transmitting end setting panel schematic diagram provided for an embodiment of the application is shown in the schematic diagram.

[0025] Figure 6 A receiving end setting panel schematic diagram provided for an embodiment of the application is shown in the schematic diagram.

[0026] Figure 7 A USRP signal acquisition program diagram provided for an embodiment of the application is shown in the schematic diagram.

[0027] Figure 8A migration training sample limited case migration learning and other scheme comparison chart provided for the embodiment of the application;

[0028] Figure 9 A migration training sample sufficient case migration learning and other scheme comparison chart provided for the embodiment of the application. DETAILED DESCRIPTION

[0029] The terms "first", "second", "third", "fourth" and the like in the description and claims of the present application, and in the above drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the use of the terms so construed herein is merely for convenience and only to aid in understanding the application and is in no way a limitation on its broader scope. It is also to be understood that the use of relational terms such as "first", "second", "third", "fourth", and the like, if any, are used solely to distinguish one from another, without necessarily implying a serial or chronological order, unless clearly indicated by the context of the description.

[0030] Modulation signal identification refers to the process of identifying the modulation method of a received signal in wireless communication. Modulation is the process of converting digital or analog signals into a form suitable for transmission over a wireless channel, while modulation signal identification is the process of determining the modulation method of a received signal by analyzing its characteristics. In wireless communication, modulation signal identification is very important for ensuring correct signal demodulation and high-quality communication, and is an important technical link in wireless communication systems.

[0031] For the sake of understanding, the specific process of the embodiment of the application is described below, and the USRP modulation signal identification method based on migration deep learning provided by the embodiment of the application specifically includes the following steps:

[0032] Step 1, signal data acquisition: wireless signals are transmitted by the transmitting end, and the USRP receiving end stores the signal received by the signal acquisition program after receiving the signal, providing the required signal training set and signal test set for the CNN model of modulation signal identification, wherein the signal training set includes a pre-training data set and a migration learning training data set;

[0033] In this embodiment, regarding the pre-training data set: the classified signal set used in the channel environment I is Ω4={BPSK,8PSK,4QAM,16QAM}, the signal sequence length is L=100, and the average signal-to-noise ratio of the sequence is uniformly distributed between [-10dB, 20dB], and the number of each type of signal is N I= 100000. The training set is set to receive N signals of each type, a total of 4N signal sequences. In training, 90% x 4N is taken as the training set, and 10% x 4N is used for cross-validation. The test set is 1000 for each signal type, with an SNR range of [-10dB, 20dB], testing every interval, and taking the average to get the modulation signal recognition rate.

[0034] Regarding the migration learning training data set: in channel environment II, the number of signals of each type in the migration learning training set is set to N II , and in the test set, there are 1000 for each signal type, with an SNR range of [-10dB, 20dB], testing every 2dB interval, and finally taking the average result to get the modulation signal recognition rate.

[0035] Step 2, pre-training of convolutional neural network model: through the generated pre-training data set, the convolutional neural network model for modulation signal recognition is pre-trained;

[0036] Step 3, migration learning training of convolutional neural network model: through the generated migration learning training data set, the convolutional neural network model for modulation signal recognition is pre-trained;

[0037] Step 4: Deploy the trained migration learning convolutional neural network model to the signal test set to verify the modulation signal recognition result.

[0038] In this embodiment, LabVIEW graphical programming is performed on the host computer to realize a transceiver system that can switch multiple modulation signals. After the USRP receiving end receives the signal, it stores the signal reception through the signal acquisition program, providing the required signal training set and signal test set for the CNN model for modulation signal recognition. Deep learning is used for modulation signal recognition, and the CNN model is trained. After the training is completed, the trained deep learning model is tested and verified. The trained modulation signal recognition module is deployed to the actual application scenario for modulation recognition of the receiver. Through the communication test environment built based on USRP software radio, it is verified that the signal modulation recognition technology based on deep learning can well solve the problem of modulation signal recognition in wireless communication.

[0039] In this embodiment, LabVIEW (Laboratory Virtual Instrument Engineering Workbench) is a graphical programming language, mainly used in data acquisition, instrument control and industrial automation, etc. LabVIEW enables engineers and scientists to quickly develop complex measurement and control systems with its unique graphical programming approach. The core of LabVIEW is a data flow-based programming model, users build programs by dragging and dropping graphical functions and structures, these graphical elements are called Virtual Instruments (VIs). LabVIEW provides a rich library and tools to support various data acquisition hardware devices, such as data acquisition cards, GPIB devices, serial devices, etc. In addition, LabVIEW also supports interfaces with other programming languages, such as C, Python, etc., allowing users to flexibly extend their functionality. Another notable feature of LabVIEW is its powerful data visualization capabilities, users can easily create various charts and graphical interfaces to display data and results in real time. This makes LabVIEW very popular in testing, measurement and control systems. In addition, LabVIEW provides rich data analysis and signal processing tools to help users process and analyze data; LabVIEW is a powerful graphical programming tool widely used in engineering, scientific research and industrial automation, etc. Its unique programming approach and powerful features enable users to efficiently develop complex measurement and control systems.

[0040] In this embodiment, the core functions of software radio, including frequency conversion, analog-to-digital conversion (A / D) and digital-to-analog conversion (D / A) and RF drive, are implemented by hardware platform USRP RIO, in addition to this, all other functions of software radio are designed and implemented through software, which means that all extended functions need to be designed and programmed in addition to the most basic receiving and transmitting functions. This design approach makes software radio highly flexible and scalable, allowing for customized development and optimization based on different application scenarios and requirements.

[0041] In this embodiment, some main physical parameters of USRP RIO are as follows: USRO RIO device has 2x2 multiple input / multiple output radio frequency RF transceiver, which can realize double transmission and double reception. Moreover, the maximum transmission signal power of the RF bandwidth transceiver is 30dBm, it can receive or transmit signals from 1.2GHz to 6GHz, the maximum instantaneous bandwidth can reach 40MHz, the maximum bidirectional point-to-point bandwidth of USRP RIO device is 830MB / s, and the digital backend interface is PCle x4 bus, its speed can reach 800MB / s.

[0042] In this embodiment, the software part of the experiment uses Labview2015 and Python3.7 programming languages to design and debug the program; Labview2015 is used to build the user interface and perform preliminary data processing, while Python3.7 is used to implement more complex algorithms and deep learning models. By using the powerful library support of Python3.7, a deep transfer learning convolutional neural network (CNN) model is designed to process and analyze signal data. In terms of signal acquisition, a universal software radio peripheral (USRP) device is used to receive wireless signals. Through USRP, the required signal data can be efficiently collected, and a detailed dataset containing the modulated signal is generated, which provides rich training samples for subsequent deep learning model training. In addition, the radio frequency transceiver driver provided by the software itself is extended based on Python3.7, and a series of functions that meet the needs of the invention are added. The expansion of these functions enables better control of the USRP device and more accurate signal acquisition and processing. Through these improvements, the software is successfully combined with the hardware platform USRP to achieve the design and implementation of wireless signal transmission communication. During the implementation of the entire invention, the advantages of Labview2015 and Python3.7 are fully utilized, and the powerful performance of the USRP hardware platform is combined to ensure the efficiency and accuracy of signal acquisition, processing, and transmission. Through this combination of software and hardware, not only is the flexibility and scalability of the experiment improved, but also a solid foundation is laid for future research and development. Please refer to Figure 1 As shown in the experimental environment, a transmitter and a receiver are set up. After the wireless communication and transceiver process of the USRP, LabVIEW collects the received modulated signal to form a dataset. Using deep learning technology, the dataset is provided to the convolutional neural network for training and learning, and offline modulated signal recognition processing is performed.

[0043] The program of the experiment mainly consists of four parts: transmitter and receiver, dataset extraction module, CNN recognition module.

[0044] Transmitter and receiver experimental setup:

[0045] During the process of data communication experiment, the transmitter and receiver play a crucial role. In order to meet the experimental requirements, a USRP (Universal Software Radio Peripheral) device is used, which has high flexibility and configurability. On the USRP device, multiple antennas can be installed and configured to achieve various communication modes. Specifically, two single antennas are configured as the transmitting antenna and the receiving antenna on this USRP device, although they are physically located on the same device, but in the experiment, the data communication between them is completed through a wireless channel. This setup enables the simulation of data transmission in a real wireless environment; in this way, the performance of wireless communication systems can be effectively simulated and tested in laboratory conditions. This self-transmitting and self-receiving SISO (Single Input Single Output) configuration not only simplifies the complexity of experimental equipment, but also provides a deep understanding of the characteristics of wireless channels. In this way, the performance of wireless communication systems can be better evaluated and optimized to ensure the desired results in practical applications.

[0046] Taking the transmitter and receiver as an example, Figure 2 is the functional block diagram of the transmitter, Figure 3 is the functional block diagram of the receiver, Figure 2 As can be seen from the transmitter, from the signal source, through source coding, channel coding, signal modulation, USRP up-conversion, digital-to-analog conversion (DAC), the RF transmission module sends the signal to the wireless channel. In order to modify the RF parameters of the transmitter in real time, such as center frequency, gain, etc., an external interface is added to the RF transmission module to realize the calling of the modification of the transmission parameter module. Figure 3 As can be seen from the receiver, from the receiving antenna, the wireless information is received, and after passing through the RF receiving module, the analog-to-digital conversion (CAD) and down-conversion processing, the modulated signal is received. The RF receiving module adds an external interface to achieve real-time modification of the RF parameters of the receiver.

[0047] Dataset extraction module settings:

[0048] Wireless signals are transmitted through the transmitter, and after the USRP receiver receives the signal, it stores the signal reception through the signal acquisition program, providing the required signal training set and signal test set for the CNN model for modulated signal recognition.

[0049] CNN recognition module settings:

[0050] Figure 4 The CNN recognition module settings are as follows:

[0051] 1. Feature extraction network design: two layers of convolutional layers are adopted, and the convolution step is 1. Two layers of convolutional layers are adopted, the first layer is a 64-kernel convolutional layer scheme, the size of the first convolution kernel is 2x4, the activation function is ReLU, a bias term is adopted and zero value initialization is performed, the weight initial method is Glorot uniform distribution initialization, and the convolution kernel does not consider the completion method, the regularization term and the weight constraint; the number of convolution kernels of the second convolutional layer is 16, the size of the convolution kernel is 1x4, and the rest of the settings are the same as the first convolutional layer.

[0052] 2. Recognition network design: the subsequent fully connected neural network includes three fully connected layers. The first two layers are set to 64 and 16 neurons respectively, and the activation function is ReLU; the weight initialization adopts the Glorot uniform distribution method, which is also called Xavier uniform distribution. In this way, the weight can maintain a small variance during initialization, so that the signal can maintain a relatively balanced propagation in each layer of the neural network. In addition, in order to prevent overfitting and ensure the generalization ability of the model, a regularization term is introduced during training. The regularization term adds an additional penalty term to the loss function to constrain the size of the weight, so as to avoid the model being too complex due to the weight being too large. Common regularization methods include L1 regularization and L2 regularization, where L2 regularization is also called weight decay. In this way, the complexity of the model can be effectively controlled, and its performance on unknown data can be improved. The output layer activation function adopts Softmax.

[0053] The training parameters of the CNN are specifically set as follows: the Adam optimization method and the multi-class cross-entropy loss function are adopted. The learning rate is set to 10 -4 , the learning rate update related parameters β1=0.9, β2=0.999; the fuzzy factor is set to 10 -8 , and the decay value of the updated learning rate is 0.

[0054] The experimental process of the embodiment is as follows:

[0055] In order to ensure the efficient operation of the wireless communication system, the signal transceiving parameters are set, the signal data is collected, and the signal data is collected, as shown in Figure 5 and Figure 6, respectively adjustable radio frequency parameters of the transmitting end and the receiving end, the radio frequency parameters of the transmitting end are set as follows: radio frequency antenna, center frequency, transmitting gain, modulation mode and the like, wherein the radio frequency antenna is to ensure that the signal can be effectively transmitted, the center frequency is a core frequency point of signal transmission, the adjustment of the transmitting gain determines the transmitting strength of the signal, and the modulation mode directly affects the transmission quality and efficiency of the signal; the radio frequency parameters of the receiving end are set as follows: radio frequency antenna, center frequency, receiving gain and the like, wherein the radio frequency antenna is to ensure that the signal can be effectively received, the center frequency needs to be consistent with the transmitting end to ensure correct reception of the signal, and the adjustment of the receiving gain determines the sensitivity and range of signal reception, wireless communication transmission of different modulation signals under different signal-to-noise ratios is realized through setting and configuration, not only the stability and reliability of the communication system are improved, but also efficient signal transmission can be maintained in various complex environments;

[0056] The user starts to prepare communication, and there are eight kinds of modulation signal types in the transmission signal between the transceivers, which are 2ASK, 4ASK, BPSK and 4PSF respectively.

[0057] Signal data collection is performed: Figure 7 For the USRP signal collection program, the wireless signal is transmitted through the transmitting end, and after the USRP receiving end receives the signal, the signal receiving is stored through the signal collection program, so as to provide the required signal training set and signal test set for the CNN model for modulation signal recognition;

[0058] CNN model pre-training is performed: the CNN model proposed in the present text for modulation signal recognition is pre-trained through the generated pre-training data set;

[0059] CNN model transfer learning training is performed: the CNN model proposed in the present text for modulation signal recognition is pre-trained through the generated transfer learning training data set;

[0060] The trained CNN model is deployed to the test set data to verify the modulation signal recognition result.

[0061] Figure 8 For the comparison between the transfer learning and other schemes under the condition that the transfer training samples are limited, the experimental results are shown in the figure. Through experimental analysis, it can be obtained that when the transfer training samples are limited, the method based on deep transfer learning is lower than the scheme based on average likelihood ratio test, but still has an advantage in recognition performance. Compared with the method of retraining CNN and the performance of the feature processing combined with the full neural network, the machine learning scheme based on the CNN deep transfer learning algorithm shows the advantage under the condition that the transfer training samples are limited, and still has relatively excellent modulation signal recognition performance under low data volume training.

[0062] Figure 9 The experimental results are shown in the figure. In the case of sufficient migration training samples, the modulation signal recognition performance of the scheme based on CNN deep migration learning and the retraining CNN is basically consistent, and the modulation signal recognition performance under high signal-to-noise ratio has approached the performance peak, that is, the traditional ALRT modulation signal recognition performance. In contrast, the scheme based on modulation signal instantaneous feature and full neural network (TF-NN) is slightly weaker than the above method. The reason is that the convolution layer structure in the CNN extracts signal features better than the instantaneous feature extraction effect.

[0063] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above embodiments, and the above embodiments and descriptions in the specification are only preferred examples of the present application and are not intended to limit the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for USRP modulated signal recognition based on migration deep learning, characterized in that, The USRP modulation signal recognition method based on migration deep learning comprises the following steps: Step 1, signal data acquisition: wireless signals are transmitted by a transmitting end, and after the USRP receiving end receives the signals, the signals are stored through a signal acquisition program, providing the required signal training set and signal test set for the CNN model for modulation signal recognition, wherein the signal training set includes a pre-training data set and a migration learning training data set; Step 2, pre-training of the convolutional neural network model: the pre-training data set is generated to pre-train the convolutional neural network model for modulation signal recognition; Step 3, migration learning training of the convolutional neural network model: the migration learning training data set is generated to pre-train the convolutional neural network model for modulation signal recognition; Step 4: deploy the trained migration learning convolutional neural network model to the signal test set to verify the modulation signal recognition result.

2. The USRP modulated signal recognition method based on migration deep learning according to claim 1, wherein, The transmitting end radio frequency parameters in step 1 include at least radio frequency antenna, center frequency, transmission gain and modulation mode, and the USRP receiving end radio frequency parameters include at least radio frequency antenna, center frequency and receiving gain.

3. The USRP modulated signal recognition method based on migration deep learning according to claim 1, wherein, In step 1, the USRP receiving end receives wireless information from the receiving antenna, and after passing through the RF receiving module, analog-digital conversion and down-conversion processing, the modulation signal is received.

4. The USRP modulated signal recognition method based on migration deep learning according to claim 1, wherein, The pre-training data set in step 1: the classified signal set adopted in the channel environment I is Ω4={BPSK, 8PSK, 4QAM, 16QAM}, the signal sequence length is L=100, the average signal-to-noise ratio of the sequence is uniformly distributed between [-10dB, 20dB], the number of each signal is N I =100000, the pre-training data set is set to receive N for each signal, a total of 4N signal sequences, 90% of 4N is taken as the training set in the training, and 10% of 4N is used for cross-validation, the test set is 1000 for each signal type, the SNR range is [-10dB, 20dB], testing is performed every interval, and the average is taken to obtain the modulation signal recognition rate.

5. The USRP modulated signal recognition method based on migration deep learning according to claim 1, wherein, The number of signals of each class in the training set of the transfer learning in step 1 is set as N II In the test set, there are 1000 signals of each type, the SNR range is [-10dB, 20dB], the test is performed every 2dB interval, and the average result is taken to obtain the modulation signal recognition rate.

6. The USRP modulated signal recognition method based on migration deep learning according to claim 1, wherein, Step 3 specifically includes: the USRP receiver collects signal data sets in two different channel environments, uses the migration learning convolutional neural network to identify the modulation signal, and migrates and trains the convolutional layer and the fully connected layer based on the convolutional neural network.

7. The USRP modulated signal recognition method based on migration deep learning according to claim 6, characterized in that, The migration learning convolutional neural network adopts two convolutional layers, the convolution step is 1, the first layer is a 64-kernel convolutional layer, the size of the first convolution kernel is 2x4, the activation function is ReLU, the bias term is used and zero value initialization is performed, the weight initial method is Glorot uniform distribution initialization, the number of convolution kernels of the second convolutional layer is 16, the size of the convolution kernel is 1x4, and the remaining settings are the same as those of the first convolutional layer.

8. The USRP modulated signal recognition method based on migration deep learning according to claim 6, characterized in that, The migration learning convolutional neural network fully connected neural network includes three fully connected layers, the first two layers are set to 64 and 16 neurons respectively, the activation function is ReLU, the weight is initialized by Glorot uniform distribution, and the regularization term and weight constraint are used, and the output layer activation function uses Softmax.

Images