Abstract
<title>Abstract</title> <p>This paper presents an FPGA-based real-time implementation of an IMU/GPS sensor fusion algorithm using an Extended Kalman Filter (EKF) synthesized through High-Level Synthesis (HLS). The proposed design targets low-latency inertial navigation by accelerating the EKF prediction and update stages on a Xilinx Kintex-7 FPGA platform. Several implementation challenges associated with direct MATLAB-to-HLS translation are addressed, including toolbox dependency removal, fixed-size matrix restructuring, and synthesizable control-flow generation. To improve hardware efficiency, iterative finite-difference Jacobian evaluation is replaced with a closed-form analytical formulation, while matrix operations are optimized using Cholesky-based Kalman gain computation and loop-level HLS directives such as PIPELINE and ARRAY_PARTITION. In addition, the original 22-dimensional MATLAB navigation model is reduced to a 16-state EKF architecture suitable for FPGA implementation. The proposed implementation was synthesized using Vitis HLS 2023.1 targeting a Xilinx Kintex-7 xc7k410t-ffg900-2 FPGA. Experimental results demonstrate successful real-time operation at 50 MHz with total estimation latency between 1.12 ms and 1.21 ms per update cycle while utilizing 191 DSP slices (12% device utilization). Numerical validation against the MATLAB reference model shows relative output errors below 10^-3%, confirming functional equivalence between software and hardware implementations.</p>