Reconfigurable hardware acceleration system for extended Kalman filtering

Abstract

The invention discloses a reconfigurable hardware acceleration system for extended Kalman filtering, and belongs to the field of hardware acceleration design of algorithms. According to the method, the matrix data buffer is designed on the basis of analyzing the data equivalence between matrix multiplication in the EKF algorithm, and when the reconfigurable PE array performs current matrix multiplication calculation, the matrix data buffer provides matrix data calculated by previous matrix multiplication for the reconfigurable PE array, so that the reusability of the matrix data is fully realized, and the reusability of the reconfigurable PE array is improved. According to the method, data migration between off-chip and on-chip is reduced, operation acceleration of all matrix multiplication in the EKF algorithm is realized, and energy consumption is reduced at the same time. The data symmetry and the data sparsity in the EKF algorithm are fully utilized, trigonometric function and root extraction calculation are rapidly achieved, the reconfigurable PE array is provided with FIFO, data interaction between modules is achieved on a chip, and other optimization means are adopted, so that the calculation speed is further increased.

Description

technical field [0001] The invention belongs to the technical field of motors, and more particularly, relates to a reconfigurable hardware acceleration system with extended Kalman filtering. Background technique [0002] The Kalman filter algorithm is an algorithm that uses the state equation and observation equation of a linear system to optimally estimate the state of the system. For nonlinear systems, the Kalman filter algorithm is no longer applicable, and the extended Kalman filter algorithm (EKF) achieves the optimal estimation of nonlinear systems by approximate linear processing of nonlinear problems. Due to its superiority in digital filtering, the EKF algorithm is widely used in algorithms such as simultaneous localization and mapping (SLAM), such as Mono SLAM, VSLAM, etc. [0003] The EKF algorithm is mainly divided into two steps, namely the prediction step and the update step. The prediction step mainly predicts the position vector and covariance of the next m...

AI Technical Summary

Problems solved by technology

The EKF accelerator published by D.T.Tertei et al. at the International Conference on Reconfigurable Computing and FPGA in 2014 only supports a floating-point calculation acceleration of matrix multiplication. Compared with fixed-point design, floating-point calculation has higher precision, but Greatly increased computing resource overhead, only one matrix multiplication can be accelerated, and the area efficiency of the accelerator is also greatly reduced

In 2016, D.T.Tertei and others published an EKF fixed-point hardware accelerator that can accelerate two matrix multiplications in the ELSEVIER Journal of Computer and Electronic Engineering, but it still cannot use the same computing hardware through reconfigurable design to complete the EKF. Computational acceleration of all matrix multiplications

Images