70 likes | 200 Views
Spring 2004 ECE 734 Course Project Tool for the Generation and Optimization of DFGs from standard filter kernals. Murugappan Senthilvelan May 4 th 2004. Motivation. Algorithmic level design space exploration is a task often reserved for human experts
E N D
Spring 2004 ECE 734 Course ProjectTool for the Generation and Optimization of DFGs from standard filter kernals Murugappan Senthilvelan May 4th 2004
Motivation • Algorithmic level design space exploration is a task often reserved for human experts • However in Digital filter domain, algorithms that systematically optimize DFGs have been proposed. • Takes the edge weight matrix and node weight matrices as input • These matrices generated manually by the designer • Need for automatic generation of DFGs from High level language code !
Filter Types Supported • Standard FIR Filter structure: • y(n) = a1 * x(n) + a2 * x(n-1) + a3 * x(n-2) • Standard IIR Filter structure: • y(n) = a1 * y(n-1) + b1 * x(n) + b2 * x(n-1) • Cascaded Filter realizations: • u(n) = a1 * x(n) + a2 * x(n-1) + a3 * x(n-2) y(n) = c1 * u(n) – c2 * y(n-1) • Parallel Filter realizations: • u(n) = a1 * x(n) + a2 * x(n-1) + a3 * x(n-2) v(n) = b1 * x(n) + b2 * x(n-1) + b3 * x(n-2) y(n) = u(n) + v(n)
Netlist Representation • Best way of representing graphs Example netlist: 1 Input 0 3 Output 0 2 Add 1 4 Mul 2 1 2 0 2 3 0 3 4 1 4 2 0
Loop Optimizations • Cutset Retiming • Takes edge weight matrix, node weight matrix and the target iteration period as input and outputs the retimed netlist • Unfolding • Takes edge weight matrix and the unfolding factor as input and outputs the unfolded netlist