Implementation of Custom Precision Floating Point Arithmetic on FPGAs

Floating point arithmetic is a common requirement in signal processing, image processing and real time data acquisition & processing algorithms. Implementation of such algorithms on FPGA requires an efficient implementation of floating point arithmetic core as an initial process. We have presented an empirical result of the implementation of custom-precision floating point numbers on an FPGA processor using the rules of IEEE standards defined for single and double precision floating point numbers. Floating point operations are difficult to implement on FPGAs because of their complexity in calculations and their hardware utilization for such calculations. In this paper, we have described and evaluated the performance of custom-precision, pipelined, floating point arithmetic core for the conversion to and from signed binary numbers. Then, we have assessed the practical implications of using these algorithms on the Xilinx Spartan 3E FPGA boards.