Research On Signal Demodulation Technology Based On Deep Learning Network

The error rate of demodulation is the key factor that affects the performance of wireless communication system.The received signal can be correctly and effectively judged by the neural network based on feature expression.In cognitive radio,intelligent communication and other fields,the receiver needs to accurately identify and demodulate the signals of different modulation types.The deep learning network can timely change the judgment rules according to the changes of modulation types.Moreover,the operation mode of the deep learning network can reduce the operation complexity of the demodulation system.Firstly,for MPSK,MQAM and MAPSK digital modulation signals,this thesis proposes a demodulation method based on SAE network.The characteristics of baseband sampling signal can be extracted by SAE network,and demodulated bit stream can be output through Soft-max classification layer.The simulation results show that in the environment of AWGN and multi-path fading channel,the performance of the demodulation method based on SAE network is 3?5dB higher than that of the traditional coherent demodulation method,and the operation complexity is lower than that of the traditional coherent demodulation method.Secondly,for the unknown modulation type signal,this thesis designs a blind demodulation system model based on a single SAE network architecture.The simulation results show that in AWGN channel environment,when the modulation type is less,the performance of demodulation method based on single SAE network is 4?5dB higher than that of traditional coherent demodulation method.Finally,for MPSK and MQAM signals,this thesis proposes a blind signal demodulation method based on modulation recognition and multi network architecture.The simulation results show that the modulation recognition method proposed in this thesis has a higher recognition accuracy than the existing methods,and can achieve high-performance demodulation of the signal.