Simultaneous Transmit‐Receive Processing for UAV Jammers: A Deep Neural Network Approach to Self‐Interference Cancellation

To address the significant residual interference fluctuations caused by the dynamic coupling between hardware nonlinearities and time‐varying channel characteristics in self‐interference (SI) signals, this paper proposes a dual‐layer SI cancellation (SIC) method based on convolutional long short‐term memory deep neural networks (CLDNN). We establish a dual‐layer cancellation model for full‐duplex jammers and derive the interference cancellation expression under the combined effects of nonlinearity and time‐varying channels. Furthermore, a CLDNN‐based network incorporating high‐order expansion terms is designed to break through the linear fitting limitations of traditional adaptive cancellation, thereby enhancing SIC performance. Simulation results confirm that the proposed dual‐layer cancellation method significantly outperforms traditional least mean squares (LMS) algorithms, convolutional neural networks (CNN), and sampled‐weight gated recurrent units (SW‐GRU) methods, achieving a 26.37 dB improvement in interference cancellation ratio (ICR).