2009 Volume 6 Issue 18 Pages 1312-1317
Although the round-off noise is normally small, it might accumulate and significantly degrade the output quality in most computationally intensive emerging applications such as FFT calculation in data communication protocols or neural networks. A new hardware-friendly error-tunable round-off method is introduced. The most important feature of this new method is that it can provide acceptable and arbitrary selectable accuracies along with different hardware implementation costs.