Research on cross domain denoising modulation recognition method for time-frequency graph style transfer
Signal modulation recognition has important applications in software radio.For low SNR ratio communication signals,the recognition rate is low.Changing the distribution of target domain data leads to a seri-ous decrease in noise reduction effect.A cross domain noise reduction modulation recognition model based on time-frequency graph style transfer is proposed to solve this problem.Firstly,the short-time Fourier transform of different modulation signals was used as input for the style transfer network for unsupervised denoising training.Then,the denoised time-frequency map was fed into the trained residual neural network for recognition,achiev-ing effective classification of communication signal modulation methods.The simulation results show that this method outperforms traditional wavelet soft thresholding image denoising methods in denoising indicators such as mean square error,peak signal-to-noise ratio,and structural similarity.At a signal-to-noise ratio of 0 dB,the av-erage recognition rate can reach 95.375%.When the signal related parameters are changed,the source and target domains of the signal change,and the denoising network still performs well,achieving cross domain denoising of the modulated signal,reflecting the strong generalization of the algorithm.
NETL
NSTL
万方数据
