Comparison of a programmable DSP and a FPGA for real-time multiscale convolution

Abstract

Real time image and signal processing is becoming increasingly important in today's military arena for functions such as tactical and strategic surveillance, surface and airborne target acquisition and tracking, self guided armaments, and remote or autonomously guided vehicles. Many application areas where a real time processing solution is required impose considerable constraints on physical size, power dissipation and cost of the solution. An additional constraint which is often imposed is that of flexibility of the solution, enabling reconfigurability or modification while in service. The possible need for reconfigurability precludes dedicated hardware solutions, which leaves as options solutions based on general purpose microprocessors, programmable digital signal processors (DSPs) and field programmable gate arrays (FPGAs). High performance microprocessors tend to dissipate too much power to be viable for low power solutions and generally require extensive glue logic to interface to the rest of the system. DSPs and FPGAs however dissipate less power yet offer high performance, and often contain logic to enable direct connection to other system components. This paper presents a comparison between the Texas Instruments TMS320C80 programmable DSP and a Xilinx FPGA for implementing a multiscale convolution (MSC) algorithm. It is beyond the scope of this paper to discuss either implementation in detail, but concentrates on the issues that have been addressed, and the merits and disadvantages of each implementation.