DDPM-based Denoising for Modulated Signal Constellation
Modulated signals are widely used in the fields of wired communication, radio communication and audio/video transmission.However, modulated signals are often interfered by noise in the transmission process, which leads to negative impact on the subsequent modulation recognition.To address this problem, this paper proposes a Denoising Diffusion Probabilistic Model ( DDPM ) based denoising method for modulated signal constellation.This method feeds the coordinates of the ideal modulation signal constellation pattern into the neural network, trains the network model based on the noise data produced by the DDPM forward diffusion denoising process, and uses the DDPM reverse diffusion process to achieve denoising for the real modulation signal constellation pattern.This method can recover modulated signal constellations with high signal-to-noise ratio under arbitrary noise interference.The experimental results show that this method exhibits significant denoising performance when dealing with Binary Phase Shift Keying ( BPSK) , Quadrature Phase Shift Keying ( QPSK) , and 8 Phase Shift Keying (8PSK) modulation methods.When the signal-to-noise ratio exceeds -5 db, the points coordinates offset of the denoised constellation is only 1.17 e-02 , and the standard deviation is only 1.53 e-04 , which can be effectively used in modulation recognition.
NETL
NSTL
万方数据
