This paper presents an analysis of challenges faced in porting an Attitude Determination and Control System (ADCS) simulation on the on-board computer of a solar-sailing nano-satellite COEPSAT-2, which is currently being developed by undergraduate students of College of Engineering Pune, with a focus on attitude estimation using the Unscented Kalman Filter. The main objective of ADCS in the mission is to orient the satellite in such a way as to maximize the solar thrust in the direction of velocity and also make communication with the ground station feasible. An accurate attitude determination and active control system is required to accomplish the task, thereby proposing sunsensors, magnetometers and gyroscope with a redundant IMU as sensors and 3-axis magnetorquers and pitch axis reaction wheel as actuators. Tri-Axial Attitude Determination (TRIAD) has been tested with sun sensors and magnetometers providing the two vectors in the satellite body frame. To tackle inherent sensor noise, which is simulated as gaussian noise, and the absence of sun sensor readings in the umbra region of the eclipse shadow of the orbit, an Unscented Kalman Filter (UKF) will be used, which will provide an attitude estimation in the umbra region and filter out noise in sensor readings in the sunlit region. Actuator limitations such as reaction wheel saturation, actuator power-current limits have been considered in the simulations. Reference orientations will be generated based on positions of the sun and the satellite, with the satellite position obtained using GPS. The RK4 propagator is used both for simulation of orbit propagation as well as a prediction model for UKF. The paper explores the important challenges of managing real time constraints while porting the ADCS algorithms on the on-board controller, optimizing the memory footprint and still keeping the attitude estimation scope intact. The control loop time constraints have been tested on Atmel's ARM-architecture microcontroller. Comparison of satellite attitude estimation techniques - Unscented Kalman Filter (UKF) and Extended Kalman Filter (EKF) - on parameters such as performance and memory footprint, alongwith simulation results of UKF filter convergence are presented. Heavy matrix operations have been optimized to match the on-board architecture to reduce running time. The comparison of estimation algorithms and time reductions on the on-board porting will be useful for other satellite missions.
