Design of RNS-KSA Based 2D FIR Filter

摘要

Two-dimensional finite impulse response filters have the benefits of phase linearity and assured stability, making them appealing in applications. 2D Finite Impulse Response (FIR) filters serve a significant role in different applications, such as image processing, medical field, pattern recognition, robot vision, and seismic or geophysical signal processing. This paper presents a novel RNS-based 2D Finite Impulse Response (FIR) filter using Residue Number System (RNS) and floating-point arithmetic operations, in which Residue Number System (RNS) provides substantially lower ranges of numbers in arithmetic computations than starting numbers and Floating-point arithmetic provides more accuracy and dynamic range than fixed-point arithmetic. The Floating Point Adder in the filter is designed using a Kogge-Stone parallel prefix adder (KSA) to improve addition speed. The proposed novel 2D FIR filter can process 32, 48, and 64-bit data by just changing the corresponding RNS moduli selection which makes it a better choice over a fixed-point-based 2D FIR filter. This filter is designed in Verilog HDL for kernel sizes 3X3, 5X5 and 7X7 and synthesized using Altera's Quartus software for FPGA Board Cyclone III (EP3C16F484C6) to evaluate the design parameters of the filter.