The rapid advancement of satellite communication systems necessitates the integration of cognitive radio technologies to ensure system adaptability. However, existing public datasets are predominantly centered on terrestrial signal features, severely constraining their applicability in space environments. Here, we introduce RML24, the first open-source benchmark dataset specifically engineered for deep learning-based satellite Telemetry, Tracking, and Command (TT&C) signal processing. We generated over 1.3 million signal samples using Software Defined Radio (SDR) and Radio Frequency (RF) transceiver platforms, encompassing 22 modulation schemes prevalent in TT&C systems. RML24 incorporates simulated satellite channel models, authentic RF link effects, and signal propagation phenomena to align the data distribution with real-world signal characteristics. All datasets and associated model codes are publicly available. Furthermore, to facilitate broader cognitive radio research, RML24 includes comprehensive and precise annotations for tasks such as deep learning-based parameter estimation and demodulation. By providing a high-quality resource tailored for satellite communications, RML24 aims to accelerate the development of intelligent, adaptive, and deep learning-driven satellite systems.
Zhang et al. (Fri,) studied this question.