Spatio‐Temporal cGAN Based Channel Estimation for Low‐Altitude Integrated Sensing and Communication Systems

ABSTRACT In low‐altitude integrated sensing and communication systems, unmanned aerial vehicles mobility induces highly time‐varying and non‐stationary channels, where pilot‐based or static‐assumption methods fail to ensure reliable channel estimation. This paper proposes a channel estimation method combining spatio‐temporal structure with conditional generative adversarial network. The U‐shaped convolutional neural network is used for channel spatial feature extraction and high‐precision reconstruction, while skip connections preserve fine‐grained details. Furthermore, a long short‐term memory network is integrated to model the temporal dynamic evolution of the channel, enhancing adaptability to non‐stationary environments. A conditional adversarial training with a PatchGAN‐based discriminator is introduced to improve local realism and distribution approximation. Simulation results confirm that the proposed method significantly improves estimation accuracy and robustness in low‐altitude scenarios.