Research on roll angle estimation method based on deep learning

Attitude measurement technology serves as a fundamental component in monitoring vehicle motion states and ensuring safety. The spinning motion of vehicles leads to a coupling of attitude angles, which significantly impacts flight control. In this paper, a deep learning approach based on the Long Short-Term Memory (LSTM) neural network is proposed to determine the real-time roll angle of a vehicle. The energy characteristics exhibited by a single antenna receiving satellite signals during the vehicle's rolling state have been analyzed in detail. A correlation between the real-time roll angle and the energy amplitude of the received signal has been established. The influence of changing satellite positions on these measurements is also discussed. Subsequently, the LSTM neural network training method is employed to extract periodic variation features from the measured signals, thereby obtaining various network parameters. These parameters are then used to predict and denoise the received signal, with the roll angle being computed by model matching. To validate the efficacy of the proposed method, a rolling experiment was conducted. The experimental results demonstrate that the LSTM-based deep learning approach effectively restores the features of the received signals, enabling accurate real-time measurement of the vehicle's roll angle.