Single-channel blind source separation and noise parameter estimation method based on U-net

The invention belongs to the technical field of communication, and particularly relates to a single-channel blind source separation and noise parameter estimation method based on U-net + +. According to the statistical physical characteristics of the mixed noise, the multi-layer feature extraction capability of the U-net + + structure is combined, the source signals are separated based on the single-channel received signals, the designed preprocessing module reduces obvious abnormal values in the received signals, the difficulty of network learning features is reduced, the risk of over-fitting is weakened, and the accuracy of the network learning features is improved. In addition, the new loss function is better matched with a mixed noise environment, the influence of different threshold noise samples on network parameters can be better described, and the complexity is further reduced through the simplified loss function. In addition, a demodulation method based on maximum likelihood can be designed in combin