Nonlinear Least-Squares State Estimation for 2D RFID-Based Motion Capture

In this paper, we present a general technique for implementing nonlinear least-squares state estimation for a two-dimensional RFID-based motion capture problem that can achieve 1.45 cm localization accuracy without the need for special initialization or tuning processes - a significant improvement over previous results in the literature. We demonstrate various algorithms that use different combinations of received signal strength (RSS), backscatter signal phase, a 3-axis accelerometer, a 3-axis gyrometer, and 3-axis magnetometer measured in a live microwave backscatter system. Permutations involving different nonlinear solvers (Gauss-Newton and Levenberg-Marquardt) and different stack levels (the number of samples in time incorporated into an estimation) are addressed.