Predicting and Clustering Attitude‐Induced Divergence in Cross‐Array AOA Measurements for LEO Satellites

Satellite geolocation of radio‐frequency emitters increasingly relies on compact angle‐of‐arrival (AOA) antenna arrays carried by small satellites. Crossed linear arrays are attractive for CubeSats, but their achievable precision is strongly orientation‐dependent and can degrade sharply in endfire‐like configurations, producing apparent ‘divergence’ of the measurement dispersion. Yet, how the spacecraft attitude maps into these critical regions has not been quantified in a mission‐design‐ready form. Here, we show an analytical attitude‐to‐performance model for cross‐array AOA receivers and a torus‐aware clustering workflow that predicts and groups attitude configurations associated with high AOA uncertainty. Our method scans the full roll‐pitch‐yaw domain, flags high‐variance samples using a quantile‐based threshold on the Cramér–Rao lower bound (CRLB), embeds periodic angles into a Cartesian space via sine‐cosine coordinates and extracts connected components as interpretable clusters. In a representative LEO CubeSat case at GNSS L1/E1, the approach reveals a finite set of divergence regions (4 clusters). These attitude ‘risk maps’ can inform offline ADCS pointing constraints and online mitigation policies, improving the reliability of AOA‐based spectrum monitoring and search‐and‐rescue missions.