Time-frequency Aliasing Modulation Signal Recognition Based On Lightweight Network

Aliasing modulation identification is an important research topic in the field of communication.Due to the complexity of the actual channel,the signal received by the receiver is often aliasing signal.Therefore,the identification of aliasing signal modulation type plays a key role in later signal processing.In this paper,experiments are carried out on different formats of aliased modulation signals.Deep learning is introduced into aliased modulation signal recognition,the recognition effect of neural network for aliased signals is studied,and the network is improved and the recognition rate is improved.The main work of this paper is as follows:1.Build the data acquisition platform and make the measured signal data set.Firstly,build a signal acquisition platform,use two USRP b210 signal generators to generate signals respectively for aliasing,and use RF sensor equipment to receive them.The measured signal considers two different types of data sets: signal-to-noise ratio data set,energy ratio data set and aliasing ratio data set.The signal-to-noise ratio data set covers the range from-20 d B to 18 db.The energy ratio and aliasing ratio data set has several classic aliasing ratios,such as 25%: 25%,50%: 50%,75%: 75%,100%: 100%,50%:100%,25%: 50%.At the same time,consider whether the aliasing signal bandwidth is the same.2.Based on the image,the aliasing signal recognition is realized by using lightweight neural network.Firstly,the principle of differential constellation recognition is introduced,and the simulation constellation data set is generated by MATLAB.After the residual network resnet18 is built,the aliasing signal constellation is recognized by its implementation.The experimental analysis is carried out for the influencing factors such as the number of network training pictures,recognition rate,aliasing ratio and energy ratio,and the problem that the algorithm is easy to be affected by frequency offset is put forward,which leads to the fourth chapter.3.Aliasing signal modulation recognition algorithm based on IQ data.Since the constellation is easily affected by frequency offset,IQ data is used instead.Therefore,the original residual network is improved.Based on it,iresnet network is obtained by adding effective information flow,introducing new mapping method and grouping convolution,which can match the size of input convolution kernel in the network model according to the characteristic dimension of input aliasing signal.In the experiment,the super parameters of neural network are demonstrated experimentally,and the best parameters are obtained.The experimental comparison and analysis are carried out for the parameters,such as signal-to-noise ratio,energy ratio,aliasing ratio,frequency offset and unknown type of aliasing signals.Experiments show that in practice,when the signal-to-noise ratio is 5d B,the recognition rate is 100%.Compared with the traditional residual network,the improved network model has better recognition effect for different aliasing situations in the environment of low signal-to-noise ratio and frequency offset.When the aliasing types considered in the training set increase,its recognition rate for unknown aliasing types is improved.