GMSK demodulation combining 1D-CNN and Bi-LSTM network over strong solar wind turbulence channel

To solve the problem of bit error rate (BER) performance degradation over strong solar wind turbulence channel, this paper addresses to Gaussian minimum frequency shift keying (GMSK) demodulation using machine learning. First, by analyzing the scintillation characteristics of the telemetry signal caused by solar wind turbulence during the solar superior conjunction, the K distribution channel model is innovatively established over strong solar wind turbulence channel. Then, the approximate probability density function of the K distribution of the established channel is studied by using Laguerre orthogonal polynomial for convenience. Additionally, the BER performance of GMSK over strong solar wind turbulence channel is analyzed. Second, using a one-dimensional convolutional neural network (1D-CNN) and bidirectional long short-term memory network (Bi-LSTM), two demodulators for GMSK are proposed. By integrating the 1D-CNN and LSTM to extract local features and process the dynamic information of GMSK signals, we propose a novel GMSK demodulator over strong solar wind turbulence channel with combining 1D-CNN and Bi-LSTM. The simulation results show that, even if the solar wind turbulence has serious effects on the signal at a low signal-to-noise ratio, the proposed combining neural network demodulation can obtain better BER performance than the classical Viterbi algorithm, demodulation using only a single 1D-CNN or Bi-LSTM. The proposed GMSK demodulator is more suitable for the deep space exploration of the solar system.