FPLD implementation and customisation in multiple target tracking applications

Abstract

Until recently, almost all methods for target tracking and estimation have followed the algorithmic path. The use of complex algorithms, such as Kalman-filter-based algorithms, generally require high speed and a high degree of parallelism, which are beyond the capabilities of available computational resources. The customisation features of field programmable logic devices (FPLDs) can be used to flexibly customise the hardware/software in FPLDs to meet the different requirements of system speed and system complex functions. FPLD-based customisable digital systems are investigated and developed to improve both performance and methodology for implementing complex multiple target tracking (MTT) algorithms. The methodology of the scalar-based direct algorithm mapping is proposed to provide a simple method to implement complex computing algorithms into the FPLD hardware circuit. The comprehensive comparison with the matrix-based systolic algorithm engineering method for systolic array implementation shows the superiority of the scalar- based direct algorithm mapping method in MTT applications. Two basic architectures for the implementation of Kalman-filter-based algorithms in FPLDs are presented. Fully- hardware-type architecture provides hardware efficiency that can reach the highest performance. Hardware/software-type architecture that includes hardware part and software part provides hardware efficiency and software flexibility. Hardware-type Kalman filter coprocessor (HKFC) and hardware/software-type Kalman filter application-specific processor (KFASP) have been designed and implemented in a single FPLD device. Their performances are two to four orders of magnitude faster than the other known implementations of the Kalman filter. As a further step in evolution, an adaptive system, FPLD-based adaptive tracking estimation computer (FATEC), is also proposed. It has several hardware-type functional units to provide the high performance as well as the adaptability to perform various functions in the adaptive MTT algorithm. The hardware/software trade-offs in its implementation have been studied and their performances have been evaluated. The criteria for optimising hardware/software trade offs have been extracted.