Most practical FPGA designs are limited to finite precision signal processing using
fixed-point arithmetic because of the cost and complexity of floating point hardware.
The vast majority of applications employ fixed point arithmetic due to is smaller
size. The key advantage of floating-point over fixed-point is its ability to automatically
scale to accommodate a wide range of values using its exponent. Floating-point is thus
preferred by programmers for non-integer computations when it is available on CPUs due
to its ease of use.
You are required to investigate the use of floating point arithmetic in FPGAs and quantify
its cost. You will first introduce both type of representations and make a comparison (from
the literature) regarding dynamic range, precision and suitability of use on FPGAs.
- The Impact of Arithmetic Representation on Implementing MLP-BP on FPGAs: A Study
- Floating-Point to Fixed-Point Transformation Toolbox
- AccelDSP Synthesis Tool Floating to Fixed Point Conversion of Matlab Algorithms
- FPGAs vs. CPUs: Trends in Peak Floating-Point Performance (PDF)
This page is maintained by Shawki Areibi, sareibi@uoguelph.ca
Last modified April 2008