欠定性时变的卫星MIMO通信系统盲分离抗干扰方法OA北大核心CSTPCD

In response to the problem of time-varying and unknown number of interfering Unmanned Aerial Vehicles(UAVs),as well as low Signal to Interference Ratio(SIR)and Signal to Noise Ratio(SNR)of Earth station observation signals in the downlink of the Multiple Input Multiple Output(MIMO)communication system of Geostationary Earth Orbit(GEO)satellites facing malicious interference from UAV clusters,this paper proposes a FPN-BiLSTM network architecture and its training method that combines Feature Pyramid Network(FPN)and Bidirectional Long Short Term Memory(BiLSTM)networks to achieve end-to-end underdetermined blind source separation with time-varying number of sources.By applying this algorithm,the expected signal bit sequence is directly extracted from underdetermined mixing observation signals without the need for estimating the number of interference sources or traditional signal demodulation.The simulation result shows that compared to other underdetermined blind source separation algorithms based on deep learning networks,the proposed algorithm has a low bit error rate and good interference cancellation performance in scenarios where the number of source signals is time-varying and SNR and SIR are as low as 0 dB and-18 dB respectively.