A hardware/software configurable architecture for Linear Quadratic Gaussian (LQG) control is presented, which is a combination of a Linear Quadratic Regulator (LQR) control law with a Discrete Kalman Filter (DKF) state estimator. LQG controllers are ideal candidates for hardware acceleration, since the DKF algorithm requires matrix inversion, which is time consuming and potentially parallelizable. In this design, a functionally equivalent DKF method, called the Sequential Discrete Kalman Filter (SDKF), is used to transform the matrix inversion into an iterative scalar inversion. The proposed design acts as an Intellectual Property (IP) Core, where the user can adjust scaling parameters in hardware and configuration parameters in software to tailor the given architecture to a wide range of physical systems. This differentiates the proposed design from other LQG implementations since it is not application specific; in fact, this architecture, which was targeted for a Xilinx Zynq-7020 FPGA, allows for systems of state size 4 to 128 and achieves a speedup of 23.6 to 167 over a 2.7GHz quad-core processor. The goal of this approach is to support a design methodology for bridging the gap between control theory and embedded systems applications. For evaluation, this architecture was compared to a pure software LQG implementation. Additionally, the approach and results of recent LQG and LQG-related hardware designs were analyzed and compared to the proposed design.
Available at: http://works.bepress.com/phillip-jones/4/