Deep Reinforcement Learning-Driven Analog Self-Interference Cancellation for LEO Navigation Augmentation Systems

Low earth orbit (LEO) navigation augmentation systems have garnered significant attention in recent years due to their ability to revolutionize global navigation services. In LEO navigation augmentation systems, the satellite constellations offer advantages such as lower latency, wider coverage, and reduced costs. However, they are significantly hindered by self-interference (SI). Traditional methods for analog self-interference cancellation (SIC) typically require a large number of taps to capture channel characteristics. To solve SI problems within limited resources, a novel analog SIC scheme is proposed based on deep reinforcement learning (DRL). This scheme employs a deep Q network (DQN) and a deep deterministic policy gradient (DDPG) unit to precisely estimate the delay and attenuation of the SI channel. Simulation results demonstrate that the proposed scheme can reduce the SI to the noise level while maintaining the comparable performance.