Research And Implementation Of Intelligent Recognition Of Modulation Signals Based On Deep Learnin

Modulation signal recognition is the basis of radio spectrum monitoring and battlefield communication reconnaissance and countermeasure,and is widely used in military and civil fields.However,in the non-cooperative communication environment,due to the complex and changeable electromagnetic environment,various communication styles and unknown signal parameters,it is very challenging to achieve efficient recognition of modulated signals.Compared with traditional modulation recognition methods,deep learning is widely used to improve the effect of modulation signal recognition with its strong self-learning ability.However,there are still problems that the recognition model consumes a lot of resources and is difficult to deploy on resource-constrained devices,and the recognition model is easily affected by the environment and is not robust.Therefore,while improving the recognition rate of modulated signals,this paper fully considers the factors such as reducing the consumption of model resources and improving the robustness of the model to study the modulation signal recognition method based on deep learning.1.Aiming at the problem of single feature selection and lack of fusion and complementarity between different features in existing modulation classification algorithms,a feature fusion method based on deep learning model is proposed.This method combines the temporal and spatial characteristics of modulated signals to obtain more distinct recognition features.The experimental results show that based on the open source data set,when the SNR is greater than 5d B,the recognition rate reaches 93.25%,and the recognition accuracy is 3%-11% higher than that based on the single feature recognition.Further use of the actual collected data for classification and recognition confirms the effectiveness of the proposed feature extraction model and fusion strategy.2.Aiming at the problem that domain drift leads to the degradation of modulation signal recognition performance and the difficulty of label labeling,a depth domain adaptive modulation recognition method is proposed.Based on the traditional deep learning model,this method adds domain adaptive components to reduce the distribution differences between domains from the feature level and the output level,so that the model can adapt to both the source domain and the target domain.The experimental results show that the average recognition rate of the unsupervised domain adaptive method proposed in this paper is 14.3%and 8.5% higher than that of the unsupervised domain adaptive method when the additive white Gaussian noise channel and Rayleigh channel are migrated.3.Aiming at the problem that the deep learning recognition model consumes large computational resources and is difficult to deploy on resource-constrained equipment,a modulation signal recognition method based on sparse deep neural network is proposed.This method uses channel sparse algorithm to construct sparse depth neural network to recognize and classify the enhanced modulated signal constellation.The experimental results show that the sparse depth neural network model can effectively reduce the model storage scale and computation amount after selecting the appropriate pruning rate,in which the model parameters are compressed by 72%,the floating point computation amount is compressed by 45%,and the recognition rate of the sparse model is 96.8%.Compared with the recognition rate of the original model of 97%,the model complexity is greatly reduced within the range of small recognition accuracy loss.4.Aiming at the problem of how to push the breakthrough key technology into engineering practice,a modulation signal recognition system based on deep learning is proposed.The system uses the vector signal transmitter VSG60 A as the signal source to transmit the modulated signal,and then USRP2974 as the receiver to collect the modulated signal,and then uses the time-frequency map extraction algorithm and the constellation blind recovery algorithm to extract the time-frequency map and constellation map features of the signal in real time in the upper computer,and finally inputs the extracted features into the designed deep learning model to complete the recognition of the modulated signal.The experimental results show that in the actual communication process,the overall recognition rate reaches 97.8% under the natural environment noise.