A Dynamic Dual Fixed-Point Arithmetic Architecture for FPGAs
Figure 17
Comparison of DDFX to DFX and double-precision floating-point solutions. (a) Euclidean distance of the DFX (plotted with blue circles) and double-precision floating point (plotted with red “+”) from the actual solution as a function of the iteration number. Note the lack of convergence of the standard DFX method. (b) Euclidean distance of DDFX (plotted with blue circles) and double-precision floating point (plotted with red “+”) from the actual solution as a function of the iteration number. The DDFX approach converges at the same rate as double-precision floating point. (c) Replot of (b) where Euclidean distance from the solution is measured in dB. The double-precision floating-point and DDFX solutions are indistinguishable in this graph. The blue circles of DDFX overlap with the red “+” of double-precision floating point. (d) PSNR of the distance from the DDFX iterate to the double-precision floating-point iterate. The high PSNR values indicate that the two solutions are virtually identical. (e) Dynamic adjustment of as a function of the iteration (, ). These results were obtained through simulations.